Computer Science Review 34 (2019) 100190

Contents lists available at ScienceDirect

Computer Science Review

journal homepage: www.elsevier.com/locate/cosrev

Systematic analysis and review of stock market prediction techniques

Dattatray P. Gandhmal, K. Kumar
School of Computer Science & Engineering, Vellore Institute of Technology, Vellore, India

article info a b s t r a c t

Article history: Prediction of stock market trends is considered as an important task and is of great attention as
Received 3 April 2019 predicting stock prices successfully may lead to attractive profits by making proper decisions. Stock
Received in revised form 2 July 2019 market prediction is a major challenge owing to non-stationary, blaring, and chaotic data, and thus,
Accepted 6 August 2019
the prediction becomes challenging among the investors to invest the money for making profits.
Available online xxxx
Several techniques are devised in the existing techniques to predict the stock market trends. This
Keywords: work presents the detailed review of 50 research papers suggesting the methodologies, like Bayesian
Stock market prediction model, Fuzzy classifier, Artificial Neural Networks (ANN), Support Vector Machine (SVM) classifier,
Bayesian model Neural Network (NN), Machine Learning Methods and so on, based on stock market prediction. The
Fuzzy classifier obtained papers are classified based on different prediction and clustering techniques. The research
ANN
gaps and the challenges faced by the existing techniques are listed and elaborated, which help the
Classification
researchers to upgrade the future works. The works are analyzed using certain datasets, software
Clustering
tools, performance evaluation measures, prediction techniques utilized, and performance attained by
different techniques. The commonly used technique for attaining effective stock market prediction is
ANN and the fuzzy-based technique. Even though a lot of research efforts, the current stock market
prediction technique still have many limits. From this survey, it can be concluded that the stock market
prediction is a very complex task, and different factors should be considered for predicting the future
of the market more accurately and efficiently.
© 2019 Published by Elsevier Inc.

Contents

1. Introduction......................................................................................................................................................................................................................... 1
2. Stock market prediction: Need and description............................................................................................................................................................. 2
2.1. Literature survey on different stock market prediction techniques................................................................................................................ 3
2.2. Classification schemes for stock market prediction .......................................................................................................................................... 3
3. Research gaps and issues .................................................................................................................................................................................................. 8
4. Analysis and discussion ..................................................................................................................................................................................................... 8
4.1. Analysis on the basis of publication years ......................................................................................................................................................... 8
4.2. Analysis based on prediction techniques............................................................................................................................................................ 8
4.3. Analysis based on clustering techniques ............................................................................................................................................................ 8
4.4. Analysis based on a software tool ....................................................................................................................................................................... 10
4.5. Analysis based on datasets ................................................................................................................................................................................... 10
4.6. Analysis based on performance metrics ............................................................................................................................................................. 11
4.7. Analysis based on values of performance metrics ............................................................................................................................................ 11
5. Conclusion ........................................................................................................................................................................................................................... 12
Declaration of competing interest.................................................................................................................................................................................... 12
References ........................................................................................................................................................................................................................... 12

1. Introduction

The advancements in stock price prediction have gained signif-

icant importance among expert analysts and investors. The stock
E-mail addresses: gdattatray@gmail.com (D.P. Gandhmal), market prediction for analyzing the trends is complicated due to
kkumar@vit.ac.in (K. Kumar). intrinsic noisy environments and large volatility with respect to

https://doi.org/10.1016/j.cosrev.2019.08.001
1574-0137/© 2019 Published by Elsevier Inc.
2 D.P. Gandhmal and K. Kumar / Computer Science Review 34 (2019) 100190

the market trends. The complexities of the stock prices adapt cer- the research works. The review is made by taking the adapted
tain factors that involve quarterly earnings’ reports, market news, methodologies, publication year, performance evaluation metrics,
and varying changing behaviors. The traders depend on vari- datasets employed, and software tools. In addition, the perfor-
ous technical indicators that are based on the stocks, which are mance attained by different techniques is used for predicting the
collected on a daily basis. Even though these indicators are used stock market trends. Here, the existing methods of stock market
to analyze the stock returns, it is complicated to forecast daily and prediction have been categorized into distinct classification and
weekly trends in the market [1]. The accurate prediction of stock clustering techniques, and further, the survey is performed for
trends is interesting and a complex task in the changing industrial the exploitation of the research gap and issues found in those
world. Several aspects, which affect the behavior of stock trends, techniques. Thus, the survey acts as the motivation for the future
are non-economic and economic factors and which are taken into extension of effective stock market prediction.
consideration. Thus, predicting the stock market is considered as The paper is structured in the following manner: Section 1
a major challenge for increasing production [2]. Traditional tech- presents a brief introduction of stock trend prediction, Section 2
niques reveal that the stock market earnings are predicted from provides needs and description regarding the prediction of the
previous stock returns and other financial variables and macroe- stock market, and literature survey of the existing stock predic-
conomics. The prediction of stock market revenues directed the tion techniques and Section 3 elucidates analysis and discussion
investors towards examining the causes of predictability. The of the existing techniques. Section 4 deliberates the research gaps
forecasting of stock trends is a difficult process as it is influenced
and issues, and the future works and Section 5 provides the
by several aspects, which involve trader’s expectations, financial
conclusion of the survey.
circumstances, administrative events, and certain aspects related
to the market trends. Moreover, the list of stock prices is usually
dynamic, complicated, noisy, nonparametric, and nonlinear by 2. Stock market prediction: Need and description
nature [3]. The forecasting of financial time series becomes an
issue due to certain complex features, like volatility, irregularities, The stock market has gained the attraction of the investors due
noise, and changing trends [4]. to advanced applications, in which the forecasting may lead to
Various models applied for predicting the stock prices are successful market prediction. The prediction of the stock trends
managed using the time series models that involve Auto- directly depends on investing and trading of stock data. The tools
Regressive Conditional Heteroscedastic (ARCH) model, General- employed for the stock market prediction can monitor, predict,
ized Auto-Regressive Moving Average (GARCH), and Auto- and regulate the market, which can be utilized for taking cor-
Regressive Moving Average (ARMA). However, these models en- rect decisions [23]. The stock market has to deal with multiple
tail historical data and hypothesis like normality postulates. Sev- industrial stock data, which covers the whole financial market.
eral methods used for stock market prediction are based on With respect to the market status, the actions are adapted by
conventional time series, such as fuzzy time series data, real the investors considering the sales and purchase. Several factors,
numbers, and design of fuzzy sets. The fuzzy time series data which influence the status of the market, are estimations of future
are implemented for stock market prediction for handling lin- incomes, a news release on profits, declaration of dividends,
guistic value data for producing precise predicting results. These management changes, and so on. The research in stock trading
methods are widely used for forecasting nonlinear and dynamic issues led to the prediction of some aspects, which are considered
datasets in the changing domains, such as tourism demand and by the experts for influencing the stock price [24]. The prediction
stock markets [5]. Many intelligent techniques, namely soft com- mechanisms based on the stock market play an important role
puting algorithms, Neural Network (NN), back propagation algo- to bring more persons and existing investors in common place.
rithm, and Genetic Algorithm (GA), are applied for predicting the The accurate results for predicting the stock market help the
stock market returns. In [6], a prediction model was designed for investors for making better decisions. The data mining tools can
predicting the stock trends with time series models. In [7], GA and
help the investors to predict future trends and behaviors and help
NN are integrated for designing hybrid expert systems to make
the institutions for yielding active solutions to make knowledge-
the investment decisions. A technique based on GA is designed
driven decisions [25]. The intelligent data analysis helps the tools
in [8], for feature discretization and determining the weights of
to generate results with ease. The extraction of useful information
Artificial Neural Networks (ANNs) [9] for predicting the index of
is an effective way for mining data [26]. Several data mining
the stock price. Time series model and NN were combined to
techniques [27] and knowledge discovery from databases [28,29]
predict the variability of the stock price in [10]. The Artificial
have been employed for analyzing the market trends. Here, the
Intelligence (AI) techniques, like ANN, were devised for predicting
data mining is essential for predicting the stock market, which
the stock market prices. Many networks used feed forward neural
networks for predicting the stock trends and evaluated multiple can search the hidden parts and increase the accuracy levels for
parametric and non-parametric models to forecast the stock analyzing the market trends using techniques, such as regression
market returns [11]. Soft computing methods are utilized to deal approach, Knowledge Discovery in Databases (KDD), fuzzy mod-
with the AI for making the decisions using the profit and loss els for making effective investment decisions [30]. Fig. 1 depicts
criterions. The techniques employed are fuzzy logic [12], Particle the functioning of the stock market prediction system.
Swarm Optimization (PSO) [13], ANN [14,15], and Support Vector Fig. 1 illustrates the block diagram of the function of the stock
Machine (SVM) [16,17]. Several researchers tried to employ fuzzy market prediction system. Initially, the historical stock data are
based techniques and randomness for optimizing the pricing collected from various datasets, such as Bombay Stock Exchange
models [18,19]. In [20], the fuzzy-based techniques are employed (BSE) dataset, Taiwan Stock Exchange Weighted Index (TAIEX),
for analyzing the market trends, and in [21], the performance of and so on, based on a technical index. Here, the technical index is
the fuzzy forecast is derived for estimating the initial values of defined as the relevance of the product to be utilized in a specific
stock price [22]. application. Then, the collected data is subjected to data pre-
The primary intention of this work is to give a detailed review processing, for removing the noise and artifacts from the datasets.
using various stock market prediction methodologies for predict- At first, the data needs attribute relevance analysis to be applied
ing future trends and stock returns. This survey considers several for removing the unwanted attributes. Then, the pre-processed
existing stock market prediction techniques that are adapted in data are employed for selecting the significant features, which
 D.P. Gandhmal and K. Kumar / Computer Science Review 34 (2019) 100190 3

Fig. 1. Block diagram of the functioning of the stock market prediction system.

can be useful for predicting the stock trends. Then, the selected 2.2. Classification schemes for stock market prediction
feature is analyzed for extracting the knowledge using a large
amount of data. The data analyzer provides improved analysis The specific research works that employ different classification
and user-friendly interface [31]. Here, the data analyzer divides schemes are elaborated in this section.
the obtained details into two classes, namely prediction details (1) Analysis based on prediction techniques
and current details, which are considered as a foundation for The analysis based on prediction techniques is elaborated in
decision making. The prediction details, the current details, and this section, which includes ANN, CNN, DSS, HMM, NN, RNN, SVM,
the threshold values are subjected to the decision-making system. SVR, and NB.
Using the threshold value, the pre-processing is done to declare
(a) ANN-based prediction techniques
profit or loss. Then, the alert notification is generated based on
ANN captures the structural relationship between a stock’s
the status. If the investor is in profit, then the shares can be used
performance and its determinant factors more accurately than
for increasing the sale, and if in loss, then this share is given more
many other statistical methods. In literature, various sets of input
attention for its development, and the NN [32] is applied to make
variables are utilized to predict stock returns. Some researchers
effective decisions.
pre-processed the input variables before applied it to the ANN for
prediction. This subsection elaborates different research works,
2.1. Literature survey on different stock market prediction tech- which employed the ANN-based stock market prediction as fol-
niques lows,
Ticknor, J.L [1] designed a model named Bayesian regularized
ANN, for predicting the behavior of the financial market. Market
This section describes the review of different stock market
trends and technical indicators are utilized as inputs for predict-
prediction techniques. Fig. 2 illustrates the categorization of dis-
ing the cost of each stock. Accurate forecasting of stock prices is
tinct stock market prediction techniques. The stock market pre- important to serve the investors to enhance the stock returns.
diction techniques are broadly categorized into two types, namely Here, the Bayesian regularized network allocates weight to the
prediction based techniques and clustering based techniques. The permitted the network for dealing with complicated models. The
techniques based on ANN, Convolutional Neural Network (CNN), model decreases the potential to improve the quality of predic-
Recurrent Neural Network (RNN), Decision Support System (DSS), tion and network generalization, but it failed to consider technical
Hidden Markov Model (HMM), Naive Bayes (NB), NN, Support indicators, which enhances the model quality for upcoming appli-
Vector Regression (SVR) and SVM are grouped under prediction cations. Rout, A.K et al. [33] developed Computationally Efficient
based techniques. Likewise, the techniques based on filtering, Functional Link Artificial Neural Network (CEFLANN) on the basis
fuzzy, k-means, and optimization are grouped under clustering of an adaptive model for predicting the financial time series data
based techniques. using Indian stock market indexes. The model adopts the Least
4 D.P. Gandhmal and K. Kumar / Computer Science Review 34 (2019) 100190

Fig. 2. Categorization of distinct stock market prediction techniques.

Mean Square (LMS) method along with the fitness function for utilized in an independent manner for optimizing the parameters
training the networks using weights. The prediction quality was of forecasting models. Zhong, X. and Enke, D [36] developed an
evaluated in terms of Mean Absolute Percentage Error (MAPE) effective mechanism based on data mining strategies to predict
using stock indexes. The inputs of the CEFLANN are selected from the daily trends of the S&P 500 Index return using the economic
the previous stocks using varying market sectors and technical features. Three techniques, which include Fuzzy Robust Principal
indicators for determining the optimal trends in stock predic- Component Analysis (FRPCA), Kernel-Based Principal Component
tion. Moreover, the optimal weights are selected by an adaptive Analysis (KPCA), and Principal Component Analysis (PCA) based
Differential Evolution (DE) mechanism. on dimensionality reduction are used for the prediction. Ad-
Shrivas, A.K. and Sharma, S.K [34] employed machine learn- vanced kernel functions along with relevant kernel parameters
ing algorithms which involve SVM, CHAID, Classification, and are required for the prediction.
Regression Technique (CART), and ANN to analyze and predict Omidi, A. et al. [37] used NN using prices of stock returns
the stock trends using BSE SENSEX data. The method lacked ad- from Iran Tractor Manufacturing Company based on the stock
vanced predictive models for predicting the stock market index, prices collected from two years. Here, NN, which is trained by
and this method was not applicable for other financial datasets, employing back propagation algorithm for making an effective
such as BSE 100, YAHOO. Chakravarty, S. and Dash, P.K [35] decision by classifying the input data. Hadavandi, E. et al. [38]
developed Functional Link Interval Type-2 Fuzzy Neural System developed an integrated method based on Genetic Fuzzy Systems
(FLIT2FNS) to predict the stock market index. This model used a (GFS) and ANN for constructing a proficient system based on
fuzzy rule base, which uses type-2 fuzzy sets in the antecedent stock price forecasting. Initially, a Stepwise Regression Analysis
part and results obtained from Functional Link Artificial Neural (SRA) was adapted for determining factors, which must influence
Network (FLANN) is in consequent part. Particle Swarm Opti- the stock prices. In addition, the raw data was split into clusters
mization (PSO), Back propagation, and learning algorithms are using the Self-Organizing Map (SOM) NN. At last, the clusters
 D.P. Gandhmal and K. Kumar / Computer Science Review 34 (2019) 100190 5

are subjected to GFS models using the rule base extraction and disrupts the original threshold value compared to other boxes.
database tuning. Patel, J et al. [39] designed a method, named The trading mechanism using two bound forecasts is built for
two-phase fusion approach, to predict the values of the future making effective decisions. The method failed to build advanced
stock market. The first phase contains SVR, and the second phase robust estimators for improving the accuracy of the forecasts by
contains Random Forest (RF), SVR, and ANN for the prediction. adopting additional soft computing methods.
The outputs generated by the model are SVR–RF, SVR-SVR, and
(d) HMM-based prediction techniques
SVR–ANN and other prediction models. However, the method
Recently, HMM is applied to forecast and predict the stock
failed to incorporate a semi-supervised system for making it more
market. HMM is successful in analyzing and predicting time
robust and the prediction to be more accurate.
depending phenomena, or time series. The different researches
Patel, J. et al. [40] designed a layer, named Trend Deterministic
adopting the HMM based stock prediction are elucidated in this
Data Preparation Layer, in prediction models for analyzing the
subsection. Badge, J., [47] developed various macro-economic
stocks. The method computes ten technical parameters utiliz- factors for Indian stock market with different macro-economic
ing the using stock trading data and represents the obtained factors, like technical indicators. These technical indicators are
technical indicators as a trend deterministic data. The method employed for deciding the patterns of the market in a specific
focused on short time prediction for estimating the values of time. There exist multiple technical indicators for yielding ef-
stock returns. Moreover, the method lacks long term prediction fective forecast by applying Principal Component Analysis (PCA).
for analyzing the quarterly stock prediction, profit returns, com- The inputs taken for the stock market prediction are selected
pany’s organizational stability, and revenues. Moghaddam, A.H technical indicators. HMM is adapted for determining the future
et al. [41] designed ANN to predict the exchange rate of daily prices and is considered as a dominant stochastic model. Gupta,
NASDAQ stocks. Several feed forward ANNs employed backprop- A., and Dhingra, B. [48] developed a Posteriori HMM approach for
agation algorithm for the training. The method takes previous predicting the stock ethics using the previous data. This method
stock prices for making the prediction. The method considers considers fractional variations in stock indexes for training the
the daily stock of NASDAQ for testing the ability of model pre- HMM. Then, HMM is utilized for making the maximum Poste-
diction. Guresen, E et al. [42] designed a model, named Gen- riori decision from the stock values. The method failed to con-
eralized Autoregressive Conditional Heteroscedasticity (GARCH), sider correlations for constructing the model. The performance
for evaluating the NN for stock market prediction. The model is improved using the quantization by taking hour-by-hour and
evaluates hybrid neural networks, Dynamic Artificial Neural Net- minute-by-minute stock values.
work (DAN2) and Multi-Layer Perceptron (MLP), for extracting
the input variables. Moreover, the method failed to discover if the (e) Naive Bayes based prediction technique
GARCH, E-GARCH provide correct effects on forecasts in terms of Naïve Bayes algorithm is a classification method, which gen-
correlated variables. erates Bayesian Networks for a given dataset based on Bayes
theorem. It presumes that the given dataset contains a particular
(b) CNN based prediction techniques feature in a class, which is unrelated to any other feature. Naïve
CNN is a feed-forward neural network. The number of hidden Bayes algorithm is easy to build and useful for very large datasets
layers in a CNN is more than that in a conventional neural and outperforms highly sophisticated classification techniques.
network. CNN is the renowned deep learning algorithms utilized The research work employed with the Naive Bayes for the stock
to predict stock markets. This subsection discusses the research market prediction is deliberated as follows. More, A.M et al. [49]
works adopting the CNN based stock market prediction as fol- used Neuro-linguistic programming (NLP) approach for obtaining
lows: Vargas, M.R et al. [43] used deep learning mechanism the stock information, and determining the stock charts, which
to predict the directional movement using Standard & Poor’s can help users to determine the right investments with high
500 index with technical indicators. The deep learning methods profit. The Hadoop framework was used for accessing a huge
are utilized for determining and analyzing complicated patterns amount of stock data parallel, and the Naive Bayes algorithm was
in the data and allow to speed up the trading process. The used for making the decisions to deal with probability interfer-
method failed to adapt reinforcement learning algorithms for ence and the method used prior knowledge for predicting the
training the model on market simulations. Zhou, X et al. [44] future trends.
developed a generic framework by adapting the LSTM and CNN
for providing training to predict frequent stock market trends. (f) NN based prediction techniques
The technique imitates the trading mode of the trader and uses NN is a series of algorithms that recognize the original re-
training and testing set for analyzing the effects of the updated lationships in a set of data through a process that mimics the
way the human brain operates. This section elaborates the re-
model cycle in predicting the performance. Moreover, the method
search works employing the NN based stock market prediction as
lacks other predictive models under multiscale conditions. Xu, B
follows: Chang, P.C [50] developed a model, named Evolving Par-
et al. [45] developed a recurrent convolutional neural network
tially Connected Neural Network (EPCNN), to forecast the stock
for predicting the stock market trend. The network captured
trends with input as technical indicators. The architecture pro-
essential information from the stock market returns using the
vides advanced features using different ANN features, in which
significant feature. In the first layer, an entity embedding layer
the connection between the neurons is randomly, and there can
was used to automate the learning process. The method failed to
be multiple layers. An evolutionary algorithm is adapted for im-
consider financial knowledge for optimizing the model for stock
proving the learning algorithm and training weights. EPCNN uses
trend prediction.
random connections between neurons for learning previous time
(c) DSS based prediction technique series data for improving the neural networks. The method failed
DSS can predict changes in stock prices, which are required by to incorporate advanced soft computing techniques for dealing
investors in the stock market. In this subsection, the stock market with other time series data. Pang, X et al. [51] developed an
prediction technique based on DSS is discussed as follows, innovative NN method to attain improved stock market predic-
Wen, Q et al. [46] developed an advanced intelligent trad- tions. The data is taken from the livestock market in real-time to
ing system using oscillation box prediction by integrating the analyze the stocks. Here, the deep LSTM based NN was designed
SVM algorithm with stock box theory. The box theory entails using the embedded layer and the LSTM neural network using
that a stock purchasing or retailing is successful if the price automatic encoder for predicting the stock trends. The embedded
6 D.P. Gandhmal and K. Kumar / Computer Science Review 34 (2019) 100190

layer and automatic encoder were used for vectorizing the data selection mechanism for retrieving essential patterns from the
using LSTM. However, the method was not applicable to the data. Xie, X.K., and Wang, H [59] designed RNN for analyzing
European and American stock market with large accuracy. the time series data. The dataset consists of the huge amount
Atsalakis, G.S et al. [52] designed a Wave Analysis Stock Pre- of intraday data from China Shanghai Shenzhen 300 Index. The
diction (WASP) system using the neuro fuzzy architecture that RNN was used to classify the daily features using intraday data.
uses certain aspects of the Elliott Wave Theory. This theory was The performance of the model was analyzed using precision and
useful for attaining accurate stock market prediction. The method average profit. Chen, W et al. [60] designed a model on the basis
was complicated as the oscillator adapted was a slow-moving of RNN with Gated Recurrent Units (GRUs) for predicting the
oscillator that was devised using the moving averages, which volatility of stocks in the Chinese stock market. The multiple price
make the scenarios difficult. Chatzis, S.P. [53] designed solid related features are subjected as an input to the model. How-
forecasting mechanism for predicting the stock market under ever, the method failed to consider advanced machine learning
varying time frames. The method integrates various machine method, like Interdependent Latent Dirichlet Allocation (ILDA),
learning mechanisms using daily stocks from a huge spectrum for making an accurate prediction.
of economies. The method applied a suite of machine learning
(h) SVM based prediction techniques
method to select the appropriate variables from large sets. Lastly,
SVM is the supervised learning model with associated learning
a bootstrap sampling was adapted to adjust the imbalanced na-
algorithms that analyze data used for classification and regression
ture of the fitting dataset. The method failed to explore deep
neural networks for setting high-frequency data with high ac- analysis. SVM is a powerful predictive tool for stock predictions
curacy. Shen, W. et al. [54] developed a Radial Basis Function in the financial market. This subsection demonstrates the stock
Neural Network (RBFNN) for training the data and predict the prediction techniques based on SVM classifiers.
stock indices using the Shanghai Stock Exchange. Artificial Fish Oztekin, A [61] designed a generic method for predicting the
Swarm Algorithm (AFSA) was employed to optimize the Radial price of daily stock movements by combining three prediction
Basis Function (RBF) in the learning process of RBF and was models, which include adaptive neuro-fuzzy inference systems,
optimized by GA and PSO. The method failed to consider non- ANN, and SVM, based on data analytics. The model was veri-
quantitative factors into mathematical algorithms for increasing fied with ten-fold stratified cross-validation for minimizing the
the accuracy of stock market prediction. random sampling bias with improved accuracy. But the method
Asadi, S et al. [55] developed a hybrid intelligent model for failed to consider other stock markets for verifying the method in
predicting the stock exchanges. The hybrid model is the inte- different countries. Porshnev, A et al. [62] designed an approach,
gration of GAs, data reprocessing mechanisms, and Levenberg– named lexicon-based approach, for improving the accuracy of
Marquardt (LM) algorithm to learn the NN. The initial weight stock market indicators using the mindsets of Twitter users. The
of NN is used for tuning the LM algorithm by adapting GA. The analysis of mindsets reveals the existence of eight crucial emo-
data pre-processing mechanisms contain input variable selec- tions in more than 755 million tweets. Moreover, SVM based
tion and data transformation for improving the overall model techniques are applied for predicting DJIA and S&P500 indicators.
accuracy. The method is tested by applying the stock exchange However, the method failed to maximize the training periods
indices for predicting the stock trends. Adebiyi, A.A. et al. [56] and enhance sentiment analysis algorithms. Ni, L.P et al. [63]
designed an upgraded predictive model for predicting the stock designed a model that combined SVM and fractal feature selec-
index using fuzzy neural network architectures. The effects of tion method for predicting stock price index. The fractal selection
utilizing hybrid technical, fundamental, and market indicators method was appropriate to solve the nonlinear issues and can
are analyzed for initiating the prediction. The input variables are spot the important features for the selection. The method failed
refined from market indicators and are subjected to the fuzzy- to consider other factors, which are responsible for testing the
neural network to enhance the accuracy for predicting the stocks. prediction model that includes micro and macro factors in pre-
However, the method failed to examine the critical impacts of dicting the stocks. Zhang, X et al. [64] designed a multi-source
expert opinion for predicting the quality of stock prices. Tsai, C.F multiple instance model that can integrate sentiments, events,
et al. [57] designed classifier ensemble mechanism for analyzing and quantifiable data into an inclusive framework. The extraction
the stock trends by examining the prediction performance. The and representation methods are used for capturing the news
ensembles are classified as follows, namely heterogeneous and events. The news events and quantitative data influence stock
homogeneous are designed. The ensembles are analyzed using fluctuations while predicting stocks market trends.
average prediction accuracy. Finally, the homogeneous classifier (i) SVR based prediction techniques
ensembles use NNs for predicting the stock returns. However, The SVR [65] utilizes the same principles as the SVM for
the method failed to consider non-economic factors and other classification, with only a few minor differences. This subsection
datasets for improving prediction accuracy. demonstrates the techniques employing the SVR based stock
(g) RNN based prediction techniques prediction system.
An RNN is a class of ANN, in which connections between nodes Kazem, A et al. [66] designed a forecasting model on the basis
form a directed graph along a temporal sequence. This allows of the firefly algorithm, SVR, and chaotic mapping for predicting
it to exhibit temporal dynamic behavior. This subsection deals prices of the stock market. The prediction model consists of
with different research works performing the stock market pre- three phases. The first phase deals with delay coordinate embed-
diction by employing the RNN are as follows: Hsieh, T.J et al. [58] ding method for reconstructing space dynamics. In the second
designed a united system, in which RNN based Artificial Bee phase, the chaotic firefly algorithm was adapted to optimize
Colony (ABC-RNN) algorithm are integrated for forecasting the the SVR parameters. Lastly, the optimized SVR is adapted to
stock prices. The model contains three phases, in which, initially, predict the prices of stock markets. Moreover, the method used
Haar wavelet is employed for decomposing the stock price time Structural Risk Minimization (SRM) in training the SVR process,
series data and to remove artifacts and noises. Secondly, the for making the method more robust than the existing models.
RNN was used for constructing the input features using Step- Kao, L.J et al. [67] designed a stock prediction model, which
wise Regression–Correlation Selection (SRCS). Thirdly, ABC was used Non-Linear Independent Component Analysis (NLICA) as
adapted for optimizing the RNN weights and biases while design- pre-processing for extracting the features from variables. The
ing the parameters. However, the method lacks advanced pattern features, named Independent Components (ICs), are subjected as
 D.P. Gandhmal and K. Kumar / Computer Science Review 34 (2019) 100190 7

the input to SVR to build the prediction models. The method if the ANFIS algorithm is able to predict the trends of the stock
can determine the concealed information from the original data market return accurately.
using the feature extraction process. Thus, NLICA helps to provide Javedani Sadaei, H. and Lee, M.H [74] developed a multilayer
valuable information for financial processing. Xiong, T et al. [68] model for predicting the stocks, including five layers. Each layer
developed an algorithm, named Firefly Algorithm based Multi- is taken into consideration for assisting forecast development by
Output Support Vector Regression (FA-MSVR) to determine the merging the problems completely. The method failed to consider
parameters of MSVR. The evaluation is done using economic the behavior of other layers discretely, which can be helpful for
criteria, cost of computation, and statistical criteria. Based on developing more layers for the systems. Chen, M.Y. and Chen,
economic criteria, the performance is evaluated based on simple B.T [5] designed an advanced fuzzy time series model for pre-
trading mechanism, by the relative forecast. Based on statistical dicting the prices of the stock market. The model was designed
criteria, the forecasting is performed using the forecast measures on the basis of granular computing approach using entropy-
and testing methods. However, the method failed to consider based discretization and binning-based partition mechanisms.
other valued time series data to forecast the tasks, which include The model adapted fuzzy time series models for predicting the
exchange rate. stock indices. The method failed to consider other essential factor,
which includes financial reports, technical indicators, and trading
(2) Analysis based on clustering techniques
volume. Wei, L.Y et al. [11] developed a system named ANFIS,
In this section, the analysis is done based on clustering tech-
which used multiple technical indicators for predicting the trends
niques developed in the literature and is described as follows, of the stock price. At first, the required technical indicators are
(a) Filtering based prediction techniques selected using advanced indicators with respect to the correlation
The filtering has attracted much attention and has a variety of matrix. Secondly, subtractive clustering mechanism is adapted,
possible applications in prediction. The distinct research works for partitioning the technical indicators using data discretization.
practicing the filtering based stock market prediction are delib- Thirdly, Fuzzy Inference System (FIS) is used for extracting the
erated in this section. Arévalo, R et al. [69] designed trading rules rules of linguistic terms from technical indicators and optimizing
using flag pattern recognition and incorporates essential innova- the FIS parameters using adaptive networks for accurate pre-
tions using past researches. Initially, a dynamic window scheme diction. However, the method was not applicable for predicting
is developed, which permits the stop loss and makes updated other stock indices. Other data discretization mechanism is re-
profit periodically. Moreover, the flag pattern follows the current quired for evaluating the variation of performances. Sadaei, H.J
trends and thus, an indicator named EMA was added for filtering et al. [75] designed an advanced fuzzy set for the fuzzy time series
the trades. The technical indicators are computed using 15-min model and was based on the trend estimator for estimating the
and 1-day timeframes that facilitate simultaneous consideration trends of data in an appropriate manner. Initially, the past data
of short and medium terms. But, the model failed to address data are fuzzified using different fuzzy sets for computing differential
snooping problems after producing certain trading rules. Ariyo, fuzzy relations. Lastly, the defuzzification is carried out to forecast
A.A et al. [70] designed a predictive model using the ARIMA model the current stock market. The imperialist competitive algorithm
for stock market prediction. The stock data published is obtained was deployed for training the model to identify future trends with
from the Nigeria Stock Exchange (NSE) and the New York Stock improved accuracy. The method failed to consider high-order
Exchange (NYSE). The results of the ARIMA models are used with differencing for dealing with seasonality.
the advanced forecasting techniques for short time prediction. (c) K-means based clustering technique
Srinivasan, P. and Ibrahim, P [71] designed a forecasting model, K-means clustering is a technique of vector quantization,
using a GARCH model. This model was adapted to forecast the which is fashionable for cluster analysis in data mining. The re-
volatility of the SENSEX index obtained from a specific period. search paper utilizing the K-means based stock market prediction
Using performance metrics, the GARCH model shows superior system is explained in this subsection: Nanda, S.R. et al. [76]
results in predicting the SENSEX Index return. designed a data mining mechanism for classifying the stocks into
(b) Fuzzy based prediction techniques clusters. Once the classification is completed, the stocks were
Fuzzy logic is a form of many-valued logic, in which the chosen from the groups for constructing a portfolio. The criterion
truth values of variables may be any real number between 0 for reducing the risk is attained by the portfolio diversification.
and 1. In this section, different researches adopting the fuzzy- The clustering mechanism divides stocks using specific invest-
based mechanisms for predicting the stock market is elucidated. ment criteria. The stock returns are used along with valuation
Esfahanipour, A. and Aghamiri, W [72] designed a Neuro-Fuzzy ratios from BSE dataset. The clustering method was introduced
Inference System for predicting the stock prices. The fuzzy model for portfolio management and for choosing the stocks to attain
applied technical indexes as input, and the resulting part is a effective frontier.
linear combination of the inputs. Fuzzy C-Means (FCM) clustering (d) Optimization based clustering techniques
was adapted to identify the set of rules. The membership function Optimization is an act of deciding as fully perfect, functional,
was described using a Gaussian function. The purpose of ANFIS or effective as possible. The researches utilizing the optimization-
is to predict the variation between stock prices using several based stock market prediction system are explained in this sub-
aspects, such as political reasons, macro-economic change, tech- section. Cheng, C.H. et al. [77] designed a hybrid forecasting
nical index, and fundamental analysis. Enke, D et al. [73] designed model that used multiple technical indicators for forecasting the
a three-phase stock market prediction system, in which the first trends of stock prices. The model employed four algorithms in
stage covers multiple regression analysis for defining the financial the hybrid model for providing effective rules to forecast and
variables. In the second phase, the differential evolution-based use extracted rules along with support values using rough set
type-2 fuzzy clustering is adapted for creating a prediction model. theory. The first was the technical indicators, which are used to
In the third phase, a Fuzzy type-2 Neural Network is adapted for predict the upcoming stock prices based on correlation matri-
performing the reasoning for predicting future stock prices. Boy- ces. The second method employed was Cumulative Probability
acioglu, M.A. and Avci, D [3] developed a model, named Adaptive Distribution Approach (CDPA) for partitioning the technical in-
Network-Based Fuzzy Inference System (ANFIS), for predicting dicators and third was Rough Set Theory (RST) algorithm for
the stock market returns. The purpose of the work is to determine extracting the linguistic rules and fourth was GA for refining
8 D.P. Gandhmal and K. Kumar / Computer Science Review 34 (2019) 100190

the extracted rules to provide better prediction accuracy. But, ANN, CNN, DSS, HMM, NN, RNN, SVM, SVR, and Naive Bayes, and
the method failed to consider other data discretization methods the clustering techniques are further classified based on filtering,
and artificial intelligence algorithms for yielding improved fore- optimization, fuzzy based techniques, and k-means techniques.
casting process. AraúJo, R.D.A., and Ferreira, T.A [4] developed Moreover, these papers are analyzed using certain factors, such as
an Evolutionary Morphological-Rank-Linear (EMRLF) mechanism publication year, datasets employed, performance metrics, soft-
for overcoming the random walk to forecast financial time se- ware tools, and the values of performance metrics. From the
ries. EMRLF contains intelligent hybrid model that integrates analysis, it is evaluated that more number of research papers
Modified Genetic Algorithm (MGA) and Morphological Rank Lin- were published in the year 2011. ANN was employed in most
ear (MRL) filter to perform an evolutionary search for mini- of the papers for predicting the market trends, whereas fuzzy
based systems are employed for clustering the stock data. The
mizing the total time lags that are capable for tuning the time
most used software tool for predicting the market trends in stock
series data in MRL filter. Then, the Least Mean Square (LMS)
prediction systems is MATLAB. The frequently used datasets for
algorithm was employed for predicting the stock trends. The
the stock market prediction are BSE, TAIEX, and S & P 500 indexes.
model was unable to regulate the distortions based on time Other datasets, such as NASDAQ, Istanbul stock exchange, Tehran
while modeling financial time series data. Yeh, C.Y et al. [2] de- stock exchange index, Citegroup and motors, Yahoo finance, Chi-
signed a two-phase multiple-kernel learning algorithm by adapt- nese stock market, National bank of Greece, and Nigeria stock
ing the gradient projection method and sequential minimal op- exchange are also employed for the stock market prediction to
timization. Based on the method, the benefits of various hyper- analyze the market trends. The commonly used performance
parameter sets are devised to improve the overall performance of metrics are Mean Absolute Percent Error (MAPE), accuracy, Root-
the system. In addition, the trial-and-error was used to determine Mean-Square Error (RMSE), Mean Squared Error (MSE), Mean
suitable hyperparameter settings. absolute error (MAE), and precision. Other metrics, such as sensi-
tivity, specificity, Mean Percentage Error (MPE), Median Relative
3. Research gaps and issues Absolute Error (MRAE), Average Relative Variance (ARV), Relative
Difference in Percentage (RDP), Root Mean Squared (RMS), Coef-
In spite of the several distinct stock market prediction ficient of multiple determination (COV), F-measure, information
methodologies, there exist certain limitations, which must be gain, F1-Score, and Mathews Correlation Coefficient (MCC) are
addressed for attaining effective stock market prediction. This used in few works, for the analysis. The most used performance
section deliberates the research gaps and issues in the different metric for analyzing the performance of stock market techniques
stock market prediction methods. is MAPE. Table 1 represents the analysis chart of the collected
papers.
The challenges faced by the NN based on stock market pre-
diction are as follows: The devised ANN was not declared as 4. Analysis and discussion
an effective scheme for predicting the stock market as the neu-
ral models cannot tolerate high computational overhead due to This section explains the analysis of the stock market predic-
large neurons contained in the hidden layer and appropriate tion based on publication year, adapted methods, datasets used,
weight adaption [33]. NN, developed in [35], performed both evaluation parameters, software tools, and accuracy.
the testing and the training at a slower rate; this affected the
prediction performance. Moreover, overfitting, trapped in local 4.1. Analysis on the basis of publication years
minima and black box technique are the drawbacks which can
be handled using NN. The obtained results of NN based stock This subsection presents the analysis based on the publication
market prediction system devised in [51] were with low accuracy years of the considered stock market prediction techniques. Fig. 3
due to the influence of the misclassification of the analogous pat- illustrates the number of research papers published in the years
from 2010 to 2018. From the 50 papers surveyed, more number
terns and the network parameters utilized were not optimized.
of works i.e. 11 research papers were published in the year 2011.
The research issue in the CNN based stock prediction method
In 2018, seven research works were developed for stock market
is that the devised CNN with the deep learning framework was
prediction.
not suitable for the highly extensive applications. The rate of
recognition accuracy achieved by CNN was comparatively poorer 4.2. Analysis based on prediction techniques
than the other state of the art prediction system for the stock
prediction [45]. The devised decision support system [46] did In this subsection, the analysis is carried based on the em-
not use the practical knowledge and techniques collected, for ployed stock market prediction techniques. The techniques em-
designing a workable stock expert system, in stock investment. ployed for effective stock market prediction is depicted in Fig. 4.
ANN required prolonged training process for designing an optimal From Fig. 4, it is noted that 29% of the works employed the ANN,
model and suffered from a lack of explanation for determining 11% of the research papers used SVM, and 9% of the works are
the solution is generated [61]. The fuzzy time series model [5] based on the SVR. The RNN is employed in 9% of the researches,
required more time and the accuracy computed by the fuzzy decision support system is employed in 3% of the works, Naive
time series model for the prediction is perpetually affected due Bayes is used in 3% and 23% of the works are based on NN, 6%
to length, and the formulation of splitting methods is very com- papers are based on HMM and remaining 8% are based on CNN.
plicated, leading to complex computational process. In SVM, a Thus, NN and ANN are mostly employed techniques for stock
feature selection method was adapted for predicting the trend market prediction.
of stock markets. The feature selection method was unable to
4.3. Analysis based on clustering techniques
point out the required number of optimal features, and thus, the
accuracy of the system is highly affected. The devised SVM and Fig. 5 presents the analysis on the basis of different clus-
NN based prediction system depend on the correlation value of tering methods employed for the effective stock market pre-
the chosen feature [63]. diction. Here, 56% of the research papers are based on Fuzzy
The table contains 50 research papers, analyzed to forecast the based system, 19% of papers used Filtering, 19% of papers utilized
trends in predicting the stock market. The techniques employed optimization techniques, and 6% of papers adapted the K-means
in the papers are categorized based on prediction and clustering clustering. Thus, it can be noticed that the fuzzy-based technique
techniques. The prediction technique is further classified based on was utilized in more research papers for stock market prediction.
 D.P. Gandhmal and K. Kumar / Computer Science Review 34 (2019) 100190 9

Table 1
Analysis chart based on prediction and clustering techniques.
Ref. no Year Performance Datasets Software tools Prediction Clustering
evaluation measures employed
1 2013 MAPE Microsoft Corp and Goldman Sachs MATLAB Bayesian based
ANN
2 2014 RMSE, MAPE BSE, IBM Corp. ANN
3 2017 Accuracy DJIA Filtering
4 2016 Accuracy, Sensitivity, BIST 100 Index SVM
and Specificity
5 2015 MSE, RMSE, MAE, TAIEX Fuzzy time
MPE series model
6 2010 Accuracy TAIEX Optimization
7 2011 RMSE TAIEX Fuzzy time
series model
8 2016 MdRAE NASDAQ, TAIEX Fuzzy time
series model
9 2013 MSE, MAPE Petrobras Company Stock Prices, Directv Group Optimization
ARV, POCID, Inc Stock Prices, and Microsoft Corporation
NMSE Stock Prices,
10 2011 RDP, RMSE TAIEX Multiple-kernel Optimization
support vector
regression
11 2010 ISE National 100 Index Fuzzy system
RMS, R2, and Cov
12 2018 MAE and BSE SENSEX dataset ANN
MAPE
13 2010 RMSE, MAPE Tehran Stock Exchange Fuzzy c means
Indexes
14 2012 MAPE Citigroup and Motors Neural network
Liquidation Company
15 2012 MAPE S&P 500, BSE, and DJIA ANN
16 2018 MSE, DA Shanghai A-shares Python Neural network
composite index
17 2010 MSE , SCC S&P500 MATLAB Decision
support system
18 2010 Intraclass inertia Capitaline Databases Plus MATLAB K-means
19 2017 SGD S&P 500 index series Tensorflow1. CNN
20 2011 Hit rate National Bank Neural network Fuzzy system
of Greece index stock
21 2016 Accuracy S&P 500 MATLAB ANN
22 2012 MAPE S&P CNX NIFTY HMM
23 2011 RMSE, MAE, DJIA, FTSE, Tokyo Nikkei-225 Index (Nikkei), RNN
MAPE and TAIEX
24 2018 G-mean, Accuracy Full in-sample dataset. Neural Network
25 2010 RMSE, MAPE SENSEX Index returns of Indian stock Filtering
market,
26 2011 AER Shanghai Stock Exchanges. Neural network
27 2011 Minimum Squared Iran Teraktorsazi Factory MATLAB ANN
Error
28 2010 MAPE Dell Corporations from the IT sector and ANN
British
airlines and Ryanair airlines from Airline
sector., and daily
the stock price of IBM.
29 2012 MAPE, POCID TSE, TEPIX data Neural network
30 2011 Accuracy NSE Matlab Neural network Fuzzy system
31 2013 Recall DJIA and S&P500 Javascript SVM
Precision
F-measure
32 2013 RMSE, MAD, MAPE, Nikkei 225 stock SVR
and DS indexes, and Shanghai Stock Exchange
Composite (SSEC)

(continued on next page)

10 D.P. Gandhmal and K. Kumar / Computer Science Review 34 (2019) 100190

Table 1 (continued).
Ref. no Year Performance Datasets Software tools Prediction Clustering
evaluation measures employed
33 2011 Accuracy TEJ Neural network
dataset
34 2014 Accuracy FTSE , S&P 500 for MATLAB SVR
the US, 100 for the UK, and Nikkei 225 for
Japan,
35 2014 MAPE, S&P, BSE, CNX Nifty and Sensex from Indian ANN
MAE, rRMSE and MSE stock markets
36 2014 Accuracy, Precision, Reliance Industries and ANN
Recall, F-measure Infosys Ltd. and CNX Nifty
37 2014 RMSE TAIEX, NASDAQ, DJIA, and S&P 500 MATLAB Fuzzy time
series
38 2018 RMSRE China stock market CNN
, DPA
39 2016 Accuracy NASDAQ stock exchange MATLAB ANN
40 2016 Precision China Shanghai Shenzhen 300 Index RNN
41 2012 MAPE Apple Inc., TATA steel, Dell HMM
Inc. and IBM Corporation
42 2011 RMSE S&P 500 Index level Fuzzy c means
43 2011 MSE and NASDAQ Stock Exchange index. ANN
MAD
44 2018 Accuracy Naïve Bayes
algorithm
45 2017 RMSEMAE, and MAPE Chinese stock market RNN
46 2014 Accuracy, Precision NYSE and NSE Filtering
47 2011 Symmetrical SSECI SVM
uncertainty,
Information Gain
48 2012 MSE National Bank shares and Microsoft daily stock MATLAB SVR
MAPE
49 2018 Accuracy and MCC Yahoo Finance CNN
50 2018 F1 Score, Accuracy Chinese dataset SVM

Fig. 3. Analysis on the basis of publication year.

4.4. Analysis based on a software tool and Tensor flow. From Table 2, it can be evaluated that the MAT-
LAB is frequently used software tool for effective stock market
prediction.
This section describes the analysis done on the basis of the
4.5. Analysis based on datasets
software tool employed in the research papers. Table 2 delib-
erates the software tools adapted for performing the effective This section elaborates the analysis carried out based on the
prediction of the stock market. The major software tools em- datasets adapted in the research works. Various datasets em-
ployed in the research papers are JavaScript, Python, MATLAB, ployed for the effective stock market prediction are depicted in
 D.P. Gandhmal and K. Kumar / Computer Science Review 34 (2019) 100190 11

Table 3
Analysis based on performance metrics.
Performance metrics Number of research papers
MAPE [1,4,33–35,38,39,47,48,50,55,58,60,66,67,72]
RMSE [2,5,11,33,39,58,60,67,72–75]
Accuracy [36,40,41,45,49,51,53,56,57,59,64,68–70,77]
MSE [4,5,39,42,46,47,51,66,72]
MAE [5,34,39,58,60,72]
MRAE [4,55]
NMSE [4,72]
Hit rate [52,56]
Recall [40,62]
Precision [40,59,62,70]

Table 4
Analysis based on the accuracy.
Accuracy range Number of research papers
50%–60% [51,64]
Fig. 4. Analysis based on prediction technique.
60%–70% [11,45,57]
70%–80% [36,49,53,61,68]
80%–90% [40,56,70]
90%–100% [69]

Table 5
Analysis based on MAPE.
MAPE range Number of research papers
< 0.5 [4,33,35,47,56,58,66,72,73]
0.5–1 [34,50,60]
1–1.5 [1,38,71]
1.5–2 [40]
2–2.5 [67]

Table 6
Analysis based on RMSE.
RMSE range Number of research papers
< 0.3 [2,33,58,71,72,74,75]
Fig. 5. Analysis based on clustering technique. 0.3–0.6 [67,77]
0.6–0.9 [60]
Table 2 0.9–1.2 [73]
Analysis based on the software tool.
Implementation tools Number of research papers
JavaScript [62]
Python [51] (a) Analysis in terms of accuracy
MATLAB [1,36,37,41,46,56,66,68,74,76] The analysis in terms of the accuracy values is elaborated in
Tensor flow [43] this section. Table 4 elaborates the analysis based on accuracy
parameter specified using five ranges as 50%–60%, 60%–70%, 70%–
80%, 80%–90% and 90% and 90%–100%. From the table, it is noted
Fig. 6. The frequently used datasets for the stock market pre- that the paper [69] attained improved accuracy with accuracy
diction are BSE, TAIEX, and S & P 500 indexes. Other datasets range 90%–100% and [40,56] and [70] had attained improved
considered are NASDAQ, Istanbul stock exchange, Tehran stock accuracy within the range 80%–90%, respectively. The minimum
exchange index, Citegroup and motors, Yahoo finance, Chinese accuracy, i.e., accuracy range within 50% to 60%, is obtained by
stock market, National bank of Greece, and Nigeria stock ex- the research papers [51] and [64].
change.
(b) Analysis in terms of MAPE
The analysis in terms of the MAPE values is described in this
4.6. Analysis based on performance metrics
section. Table 5 elaborates the analysis based on MAPE parameter
with five ranges as 0.0–0.5, 0.5–1.0, 1.0–1.5, 1.5–2, and 2–2.5,
The analysis carried out in terms of performance metrics for respectively. From the table, it is evaluated that the papers [1,40,
predicting the stock market is elaborated in this section using Ta- 67,71] and [38] had attained improved MAPE within the range
ble 3. The commonly used performance metrics are MAPE, RMSE, 2–2.5, 1.5–2 and 1–1.5, respectively.
accuracy, MSE, MAE, and precision. Other metrics include sen-
sitivity, specificity, MPE, MRAE, ARV, RDP,RMS,COV, F-measure, (c) Analysis in terms of RMSE
information gain, F1-Score, and MCC, which are employed in The analysis in terms of RMSE values is described in this
research papers [69], [69], [5], [75], [75], [2], [3], [3], [62], [63], section. Table 6 elaborates the analysis in terms of RMSE param-
[64], and [45], respectively. eter with four ranges as 0.0–0.3, 0.3–0.6, 0.6–0.9, and 0.9–1.2,
respectively. From the table, it is elucidated that the papers [2,33,
4.7. Analysis based on values of performance metrics 58,71,72,75] and [74] had attained lower RMSE within the range
< 0.3.
The analysis on the basis of performance metrics value is (d) Analysis in terms of MSE
described in this section. The analysis in terms of accuracy, MAPE, The analysis in terms of the MSE values is described in this
RMSE, and MSE is elaborated in the subsection. section. Table 7 elaborates the analysis in terms of MSE parameter
12 D.P. Gandhmal and K. Kumar / Computer Science Review 34 (2019) 100190

Fig. 6. Analysis based on datasets employed.

Table 7 References
Analysis based on MSE.
MSE range Number of research papers [1] J.L. Ticknor, A Bayesian regularized artificial neural network for stock
< 0.007 [2,35,45,46] market forecasting, Expert Syst. Appl. 40 (14) (2013) 5501–5506.
0.007–1 [5,37] [2] C.Y. Yeh, C.W. Huang, S.J. Lee, A multiple-kernel support vector regression
1–2 [39,42] approach for stock market price forecasting, Expert Syst. Appl. 38 (3)
(2011) 2177–2186.
[3] M.A. Boyacioglu, D. Avci, An adaptive network-based fuzzy inference
system (ANFIS) for the prediction of stock market return: the case of the
Istanbul stock exchange, Expert Syst. Appl. 37 (12) (2010) 7908–7912.
with three ranges as < 0.007, 0.007–1, and 1–2, respectively. [4] R.D.A. AraúJo, T.A. Ferreira, A morphological-rank-linear evolutionary
From the table, it is elucidated that the papers [2,35,46] and [45] method for stock market prediction, Inform. Sci. 237 (2013) 3–17.
had attained lower RMSE within the range < 0.007. [5] M.Y. Chen, B.T. Chen, A hybrid fuzzy time series model based on granular
computing for stock price forecasting, Inform. Sci. 294 (2015) 227–241.
[6] T. Kimoto, K. Asakawa, M. Yoda, M. Takeoka, Stock market prediction
5. Conclusion system with modular neural network, in: Proceedings of the International
Joint Conference on Neural Networks, 1990, pp. 1–6.
[7] C. Nikolopoulos, P. Fellrath, A hybrid expert system for investment
This paper presented a survey of different techniques em- advising, Expert Syst. 11 (4) (1994) 245–250.
ployed for attaining effective stock market prediction, which is [8] K. Kim, I. Han, Genetic algorithms approach to feature discretization in
artificial neural networks for prediction of stock index, Expert Syst. Appl.
categorized on the basis of prediction techniques and clustering 19 (2000) 125–132.
techniques. The purpose of this survey is to classify the existing [9] P. Vijaya, G. Raju, Santosh Kumar Ray, Artificial neural network-based
techniques in terms of publication years, methodologies adapted, merging score for Meta search engine, J. Comput. Sci. 12 (7) (2016)
323–340.
datasets used, performance metrics and implementation tools,
[10] T.H. Roh, Forecasting the volatility of stock price index, Expert Syst. Appl.
using 50 research papers. The techniques utilized for the stock 33 (2007) 916–922.
market prediction involves ANN, SVM, SVR, HMM, NN, fuzzy [11] L.Y. Wei, T.L. Chen, T.H. Ho, A hybrid model based on adaptive-network-
based techniques, K-means, and so on. In addition, the research based fuzzy inference system to forecast Taiwan stock market, Expert Syst.
Appl. 38 (11) (2011) 13625–13631.
gaps and the issues for predicting the stock market are elaborated [12] M.R. Hassan, A combination of hidden Markov model and fuzzy model for
for suggesting effective future scope. The commonly used tech- stock market forecasting, Neurocomputing 72 (16–18) (2009) 3439–3446.
nique for attaining effective stock market prediction is ANN and [13] R. Majhi, G. Panda, G. Sahoo, A. Panda, A. Choubey, Prediction of S & P
500 and DJIA stock indices using particle swarm optimization technique,
the fuzzy-based technique. These techniques can be effectively
in: Proceedings of the Proceeding of the IEEE Congress on Evolutionary
utilized for controlling and monitoring the entire stock market. Computation, 2008, pp. 1276–1282.
The major challenge faced by the stock price prediction systems [14] Y. Kara, M. AcarBoyacioglu, O.K. Baykan, Predicting direction of stock
is that most of the existing techniques cannot be detected using price index movement using artifcial neural networks and support vector
machines: The sample of the Istanbul Stock Exchange, Expert Syst. Appl.
historical stock data as they are affected due to certain factors, 38 (5) (2011) 5311–5319.
which involve government policy decisions, market sentiments [15] M. Ghiassi, J. Skinner, D. Zimbra, Twitter brand sentiment analysis: a hybrid
and so on. Thus, the data from different sources are required system using n-gram analysis and dynamic artifcial neural network, Expert
Syst. Appl. 40 (16) (2013) 6266–6282.
for making decisions, and the data pre-processing is a complex
[16] W. Huang, Y. Nakamori, S.-Y. Wang, Forecasting stock market movement
task for data mining. These are major limitations that need to direction with support vector machine, Comput. Oper. Res. 32 (10) (2005)
be addressed in the future by adapting advanced stock market 2513–2522.
prediction techniques. In the future, we will try to develop the [17] A.F.S. Elsir, H. Faris, A comparison between regression, artifcial neural
networks and support vector machines for predicting stock market index,
method for providing the accurate prediction of the stock market. Int. J. Adv. Res. Artif. Intell. 4 (7) (2015).
[18] S.S. Appadoo, Pricing Financial Derivatives with Fuzzy Algebraic Models: A
Teoretical and Computational Approach, University of Manitoba, Winnipeg,
Declaration of competing interest 2006.
[19] A. Tavaneswaran, K. Tiagarajah, S.S. Appadoo, Fuzzy coefcient volatility
models with applications, Math. Comput. Modelling 45 (7–8) (2007)
The authors declare that they have no known competing finan-
777–786.
cial interests or personal relationships that could have appeared [20] C. Carlsson, R. Fuller, On possibilistic mean value and variance of fuzzy
to influence the work reported in this paper. numbers, Fuzzy Sets and Systems 122 (2) (2001) 315–326.
 D.P. Gandhmal and K. Kumar / Computer Science Review 34 (2019) 100190 13

[21] A. Tavaneswaran, S.S. Appadoo, A. Paseka, Weighted possibilistic moments [49] A.M. More, P.U. Rathod, R.H. Patil, D.R. Sarode, B.E. Student, Stock market
of fuzzy numbers with applications to GARCH modeling and option pricing, prediction system using hadoop, Internat. J. Engrg. Sci. (2018) 16138.
Math. Comput. Modelling 49 (1–2) (2009) 352–368. [50] P.C. Chang, A novel model by evolving partially connected neural network
[22] X. Zhou, Z. Pan, G. Hu, S. Tang, C. Zhao, Stock market prediction on for stock price trend forecasting, Expert Syst. Appl. 39 (1) (2012) 611–620.
high-frequency data using generative adversarial nets, Math. Probl. Eng. [51] X. Pang, Y. Zhou, P. Wang, W. Lin, V. Chang, An innovative neural network
(2018). approach for stock market prediction, J. Supercomput. (2018) 1–21.
[23] Manna Majumder, M.D. Anwar Hussian, Forecasting of Indian Stock Market [52] G.S. Atsalakis, E.M. Dimitrakakis, C.D. Zopounidis, Elliott Wave Theory and
Index using Artificial Neural Network, National Stock Exchange of India neuro-fuzzy systems, in stock market prediction: The WASP system, Expert
Limited, 2009. Syst. Appl. 38 (8) (2011) 9196–9206.
[53] S.P. Chatzis, V. Siakoulis, A. Petropoulos, E. Stavroulakis, N. Vlachogiannakis,
[24] M.M.P. Bharambe, S.C. Dharmadhikari, Stock market analysis based on
Forecasting stock market crisis events using deep and statistical machine
artificial neural network with big data, in: Proceedings of 8th Post
learning techniques, Expert Syst. Appl. 112 (2018) 353–371.
Graduate Conference for Information Technology, 2017.
[54] W. Shen, X. Guo, C. Wu, D. Wu, Forecasting stock indices using radial basis
[25] U. Fayyad, G. Piatetsky-Shapiro, P. Smyth, The KDD process for extracting
function neural networks optimized by artificial fish swarm algorithm,
useful knowledge from volumes of data, Commun. ACM 39 (11) (1996)
Knowl.-Based Syst. 24 (3) (2011) 378–385.
27–34.
[55] S. Asadi, E. Hadavandi, F. Mehmanpazir, M.M. Nakhostin, Hybridization
[26] A. Kumar, Artificial neural networks for data mining, IASRI, Library Avenue,
of evolutionary Levenberg–Marquardt neural networks and data pre-
Pusa, New Delhi Google Scholar.
processing for stock market prediction, Knowl.-Based Syst. 35 (2012)
[27] D. Enke, S. Thawornwong, The use of data mining and neural networks for 245–258.
forecasting stock market returns, Expert Syst. Appl. 29 (4) (2005) 927–940. [56] A.A. Adebiyi, C.K. Ayo, S.O. Otokiti, Fuzzy-neural model with hybrid market
[28] U.M. Fayyad, D. Haussler, P.E. Stolorz, KDD for science data analysis: Issues indicators for stock forecasting, Int. J. Electron. Finance 5 (3) (2011)
and examples, in: KDD, 1996, pp. 50–56. 286–297.
[29] PhichhangOu, Prediction of stock market index movement by ten data [57] C.F. Tsai, Y.C. Lin, D.C. Yen, Y.M. Chen, Predicting stock returns by classifier
mining techniques, Mod. Appl. Sci. 3 (12) (2009). ensembles, Appl. Soft Comput. 11 (2) (2011) 2452–2459.
[30] A. Sharma, D. Bhuriya, U. Singh, Survey of stock market prediction using [58] T.J. Hsieh, H.F. Hsiao, W.C. Yeh, Forecasting stock markets using wavelet
machine learning approach, in: Proceedings of International Conference on transforms and recurrent neural networks: An integrated system based
Electronics, Communication and Aerospace Technology, Vol. 2, 2017, pp. on artificial bee colony algorithm, Appl. Soft Comput. 11 (2) (2011)
506–509. 2510–2525.
[31] J.R. Woolridge, C.C. Snow, Stock market reaction to strategic investment [59] X.K. Xie, H. Wang, Recurrent neural network for forecasting stock market
decisions, Strateg. Manage. J. 11 (5) (1990) 353–363. trend, in: Proceedings of International Conference on Computer Science,
[32] S. Arun Joe Babulo, et al., Stock market indices prediction with various Technology and Application, 2017, pp. 397–402.
neural network models, Int. J. Comput. Sci. Mobile Appl. 2 (3) (2014) [60] W. Chen, Y. Zhang, C.K. Yeo, C.T. Lau, B.S. Lee, Stock market predic-
42–46. tion using neural network through news on online social networks, in:
[33] A.K. Rout, B. Biswal, P.K. Dash, A hybrid FLANN and adaptive differ- Proceedings of International Conference on Smart Cities, 2017, pp. 1–6.
ential evolution model for forecasting of stock market indices, Int. J. [61] A. Oztekin, R. Kizilaslan, S. Freund, A. Iseri, A data analytic approach to
Knowl.-Based Intell. Eng. Syst. 18 (1) (2014) 23–41. forecasting daily stock returns in an emerging market, European J. Oper.
[34] A.K. Shrivas, S.K. Sharma, A Robust Predictive Model for Stock Market Index Res. 253 (3) (2016) 697–710.
Prediction using Data Mining Technique, 2018. [62] A. Porshnev, I. Redkin, A. Shevchenko, Machine learning in prediction of
[35] S. Chakravarty, P.K. Dash, A PSO based integrated functional link net and stock market indicators based on historical data and data from twitter
interval type-2 fuzzy logic system for predicting stock market indices, Appl. sentiment analysis, in: Proceedings of 13th International Conference on
Soft Comput. 12 (2) (2012) 931–941. Data Mining Workshops, 2013, pp. 440–444.
[63] L.P. Ni, Z.W. Ni, Y.Z. Gao, Stock trend prediction based on fractal feature
[36] X. Zhong, D. Enke, Forecasting daily stock market return using
selection and support vector machine, Expert Syst. Appl. 38 (5) (2011)
dimensionality reduction, Expert Syst. Appl. 67 (2017) 126–139.
5569–5576.
[37] A. Omidi, E. Nourani, M. Jalili, Forecasting stock prices using financial
[64] X. Zhang, S. Qu, J. Huang, B. Fang, P. Yu, Stock market prediction via
data mining and Neural Network, in: Proceedings of 3rd International
multi-source multiple instance learning, IEEE Access (2018).
Conference on Computer Research and Development, Vol. 3, 2011, pp. [65] C.A. Ghuge, Sachin D. Ruikar, V. Chandra Prakash, Support vector regression
242–246. and extended nearest neighbor for video object retrieval, Evol. Intell.
[38] E. Hadavandi, H. Shavandi, A. Ghanbari, Integration of genetic fuzzy (2018) 1–14.
systems and artificial neural networks for stock price forecasting, [66] A. Kazem, E. Sharifi, F.K. Hussain, M. Saberi, O.K. Hussain, Support vector
Knowl.-Based Syst. 23 (8) (2010) 800–808. regression with chaos-based firefly algorithm for stock market price
[39] J. Patel, S. Shah, P. Thakkar, K. Kotecha, Predicting stock market index using forecasting, Appl. Soft Comput. 13 (2) (2013) 947–958.
fusion of machine learning techniques, Expert Syst. Appl. 42 (4) (2015) [67] L.J. Kao, C.C. Chiu, C.J. Lu, J.L. Yang, Integration of nonlinear indepen-
2162–2172. dent component analysis and support vector regression for stock price
[40] J. Patel, S. Shah, P. Thakkar, K. Kotecha, Predicting stock and stock price forecasting, Neurocomputing 99 (2013) 534–542.
index movement using trend deterministic data preparation and machine [68] T. Xiong, Y. Bao, Z. Hu, Multiple-output support vector regression with
learning techniques, Expert Syst. Appl. 42 (1) (2015) 259–268. a firefly algorithm for interval-valued stock price index forecasting,
[41] A.H. Moghaddam, M.H. Moghaddam, M. Esfandyari, Stock market index Knowl.-Based Syst. 55 (2014) 87–100.
prediction using artificial neural network, J. Econ. Financ. Adm. Sci. 21 [69] R. Arévalo, J. García, F. Guijarro, A. Peris, A dynamic trading rule based on
(41) (2016) 89–93. filtered flag pattern recognition for stock market price forecasting, Expert
[42] E. Guresen, G. Kayakutlu, T.U. Daim, Using artificial neural network Syst. Appl. 81 (2017) 177–192.
models in stock market index prediction, Expert Syst. Appl. 38 (8) (2011) [70] A.A. Ariyo, A.O. Adewumi, Ayo, Stock price prediction using the ARIMA
10389–10397. model, in: Procedings of Computer Modelling and Simulation, 2014, pp.
[43] M.R. Vargas, B.S. de Lima, A.G. Evsukoff, Deep learning for stock market 106–112.
prediction from financial news articles, in: Proceedings of IEEE Interna- [71] P. Srinivasan, P. Ibrahim, Forecasting stock market volatility of BSE-30
tional Conference on Computational Intelligence and Virtual Environments index using GARCH models, Asia Pac. Bus. Rev. 6 (3) (2010) 47–60.
for Measurement Systems and Applications, 2017, pp. 60–65. [72] A. Esfahanipour, W. Aghamiri, Adapted neuro-fuzzy inference system on
[44] X. Zhou, Z. Pan, G. Hu, S. Tang, C. Zhao, Stock market prediction on indirect approach TSK fuzzy rule base for stock market analysis, Expert
high-frequency data using generative adversarial nets, Math. Probl. Eng. Syst. Appl. 37 (7) (2010) 4742–4748.
[73] D. Enke, M. Grauer, N. Mehdiyev, Stock market prediction with multiple
(2018).
regression, fuzzy type-2 clustering and neural networks, Procedia Comput.
[45] B. Xu, D. Zhang, S. Zhang, H. Li, H. Lin, Stock market trend prediction
Sci. 6 (2011) 201–206.
using recurrent convolutional neural networks, in: Proceedings of CCF
[74] H. JavedaniSadaei, M.H. Lee, Multilayer stock forecasting model using fuzzy
International Conference on Natural Language Processing and Chinese
time series, Sci. World J. (2014).
Computing, 2018, pp. 166–177. [75] H.J. Sadaei, R. Enayatifar, M.H. Lee, M. Mahmud, A hybrid model based on
[46] Q. Wen, Z. Yang, Y. Song, P. Jia, Automatic stock decision support system differential fuzzy logic relationships and imperialist competitive algorithm
based on box theory and SVM algorithm, Expert Syst. Appl. 37 (2) (2010) for stock market forecasting, Appl. Soft Comput. 40 (2016) 132–149.
1015–1022. [76] S.R. Nanda, B. Mahanty, M.K. Tiwari, Clustering Indian stock market data
[47] J. Badge, Forecasting of indian stock market by effective macro-economic for portfolio management, Expert Syst. Appl. 37 (12) (2010) 8793–8798.
factors and stochastic model, J. Stat. Econom. Methods 1 (2) (2012) 39–51. [77] C.H. Cheng, T.L. Chen, L.Y. Wei, A hybrid model based on rough sets theory
[48] A. Gupta, B. Dhingra, Stock market prediction using hidden markov models, and genetic algorithms for stock price forecasting, Inform. Sci. 180 (9)
in: Proceedings of Students Conference on Engineering and Systems, 2012, (2010) 1610–1629.
pp. 1–4.