Mobile App Success Prediction
Mobile App Success Prediction
Mobile App Success Prediction
http://doi.org/10.22214/ijraset.2020.6273
International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.429
Volume 8 Issue VI June 2020- Available at www.ijraset.com
Abstract: Google play store has thousands of apps and several thousand apps are added to its list every day. The competition is so
high that it is difficult for a developer to find out whether he is working in the direction of making a successful app or not. The
success of an app can be determined by factors like ratings, number of installs and reviews. In this paper we have applied
Exploratory Data Analysis to discover relationships among features of an app to predict which app will succeed. Data from
google play store was used to train three different models to predict the success of the app- Random Forest, Support Vector
Machine and Linear Regression.
Keywords: Mobile app, Google Play Store, Apple App Store, Exploratory Data, Support Vector Machine, Random Forest and
Linear Regression
I. INTRODUCTION
App Stores are not a forgotten achievement as Apple App store and Google Play Store came merely a decade back in 2008. Apps
were originally intended for productivity assistance such as email, calendar and contact databases, but the public demand for apps
caused rapid expansion into other areas such as mobile games, factory automation, GPS and location-based services, order-tracking,
and ticket purchase, so that there are now millions of apps available.[1] These app stores provide us with numerous amounts of
entertaining facilities with just few clicks in our Pcs. From that time till now we have only seen the rise in the demands of the apps
these stores provide to users. Digital technology market is greatly influenced by mobile apps growth over the years. Mobile
applications are enormous and profitable business. Looking at the market of mobile apps we can see that there is a huge difference
between the success rate of iPhone mobile apps and google mobile apps. Studies show that the iPhone has over 1.84 million apps in
their store and Google play Store has over 2.57 million apps, as of 2019. Looking at the figures, we can conclude that there is a huge
difference in numbers of apps between the two. On the other side, even though there are less iPhone users compared to Android
users, Apple stores still make 75% more revenue than the Google Play Store, as studies show in 2016. Resting aside the revenue
factor, we observe that Google App Store has always had a greater number of downloads, popularity, apps available, more free apps,
and different genres of apps. With the growth of the mobile app market, the number of app developers has also significantly
increased. With each passing day, the demand for more advanced and updated apps keeps on rising and the number of apps
available in the app stores increase by thousands. This many apps confuse the user into considering which app they can use.
Developers too loose track and only one question remains, what will make an app a success? Creating and developing apps without
considering current trends and previous data only builds up the number of apps available and burns a hole in the pocket of the
developer. We have made it a little simpler for the users and the developers, to consider which or what app to use and what kind of
apps they can deploy, respectively. Developers can see and understand what changes or advances they can bring in their ideas and
applications that will help them give a more efficient app for users and a successful app for them.
In this study, features extracted from the data made available on Google’s Play Store website is used as input to different models.
Each model understands the features of a given application, and interesting observations about their behaviour are discussed. In the
next step, we filter the data and remove the factors that can result in ambiguity or errors. We came across a paper [2] that explained
what causes an app to fail and what are bad apps. So, we perform Exploratory Data Analysis to understand the features of apps that
result in better success of an app. We use machine learning models like Random Forest Model, Support Vector Machine and Linear
Regression Model to find the correlation and understand the data.
There have been some seminal studies [2][3][4][5]. We hope to have brought some advancement in the field for the upcoming
research.
II. LITERATURE REVIEW
In paper [4], the author mined and processed the data available by the Google Play Store. He extracted the features available and
used three models to come to final conclusions. He also trained the data under those models. Using the key features, number of
installations and average user rating, allowed the author to come closer to predictions. The author used the linear model for
prediction of the average rating for the whole presented data. Principal component analysis, by using inputs of Generalized Linear
Model and Linear Regression, presented strong patterns between the features.
The author worked on the description presented by the creators of the application that presented the conclusion that about thirty-five
percent of all successful applications contain the stem “photo” and thirty-one percent of all successful applications contain the steam
“share” somewhere in their description. Successful in concluding what genres of applications interest most of the users. For his
Future Work author suggested using revenue to predict and for success metrics.
In another paper [14], the author reviews the ratings for the applications. It shines light on the fact that how store-ratings after
reaching a certain value does not affect the overall store rating even after users rate it. But on the contrary the following above is not
observable in the store-rating. It also highlights that an updated app rating depends on their versions. Thus, came the idea of rating
the versions. Version ratings can be calculated by calculating the available store-rating. The idea behind it was to help the
developers develop an app even when it reached a prominent value. They share and advise the idea of displaying the current version
ratings for the app. Similarly, in paper [13], the author works around the feature- review. The author quotes the unfruitful
combination of reviews and rating. The author furthermore works on this and builds multiple systems that detects the inconsistency
between the above two features. The system mainly revolves around the approaches of Naïve Bayes Classifier, Decision Tree,
Decision Stump, Decision Table and other few Machine Learning Algorithms and Deep Learning Approaches. They held several
surveys in the process to learn the opinion of the user and also the developers regarding the above two features. Concluding the likes
of the author, the end users and developers agreed to the idea of matched relevancy between the review and the rating of the app.
Following the research around the same sentiment analysis, in this paper [15], provides an understanding that there is a difference
between the starred ratings on the app store and the reviews by the users. So, the author proposes a new rating system which might
remove the ambiguity and difference created by the user between the reviews and the ratings. The author states that the user is
interested in downloading an app according to their rating. The author sums up the problem in 2 parts- the ambiguity and the biases.
The author's proposed system will perform a sentiment analysis on the descriptive review by the user and then generates a numeric
rating. In this way the final rating generated will be the average of the ratings. This proposed system is said to reduce the confusion
of the users and allow them to have a final rating based on both review and star rating. This paper concludes a very strong
relationship between the review and the ratings and works to help the user get the better and best app according to their likes.
In another paper [17], the authors bring up the question of why an app is not desired by the user and answers the reason behind their
failure. It can be observed and seen that not all reviews can be treated and considered under the study. Such reviews create noise and
add up to the number of errors. The authors worked to remove such reviews resulting in reduction of noise in the dataset and
providing a better performance in sentiment analysis. It results in generating more accurate polarity value. Li proposed WisCom, a
system that is able to analyse at least ten million of user ratings and comments in the app markets on three different levels. This
system is said to be firstly accomplished in identifying inconsistency in reviews and then looked for the reasons why people do not
like a particular app and how the reviews change over time according to user demands. Their proposed system and the valued
research provide an overwhelming walk through the most fascinating concerns. It generates techniques for summarizing and mining
the reviews, which will help the end users to choose the best app that too without going through the previous user descriptive
reviews. This paper allows the user to understand what went wrong in the demand or the usefulness of the app and what features and
areas they might consider to update their app to reach out to more user and move their app ranking to a higher place.
Questioning the failure same as above, in the paper [16], the authors comment on the idea of reviews and consider it the more
quantitative value and that combination of star rating with the text review will result in better quantitative estimation of service
satisfaction rating. Their average rating for the dataset was 3.6 so the author placed the rating above 3.6 to positive sentiment and
below it to negative sentiment. He used some common and popular Machine Learning Algorithms like Naïve Bayes and SVM along
with the most unusual but effective learning algorithm consisting of two processes: capturing semantic similarities and modelling
after sentiment. He used seven different training set sizes that predicted ratings with the respective sentiment.
All the papers mentioned above helped us to get all the ideas that we incorporated in our work, which made our work more
organized and eventually helped us to produce a better analysis of the features that are present in our dataset.
C. Linear Model
Linear regression is a linear approach to modeling the relationship between a scalar response (or dependent variable) and one or
more explanatory variables (or independent variables). The case of one explanatory variable is called simple linear regression. For
more than one explanatory variable, the process is called multiple linear regression. [10] It is also called the Linear Regression
Model that assumes a linear relationship between the input variables (x) and the single output variable (y). More specifically, that y
can be calculated from a linear combination of the input variables (x).
E. XGBoost Algorithm
This algorithm is considered to be dominating in now times. It is a further implementation of the gradient boosted decision trees
designed for speed and performance. The algorithm is created by Tianqi Chen. It comes under the hood of the Distributed Machine
Learning Community. XGBoost has model features like: Gradient Boosting, Stochastic Gradient Boosting and Regularized Gradient
Boosting. Its system features are: Parallelization, Distributed Computing, Computation of very large dataset that don’t fit memory
and Cache Optimization. This algorithm results in giving a good execution speed and astonishing model performance. This model is
also capable of building random forests.
VIII. CONCLUSIONS
We draw the conclusion that feature like Rating and Content Rating results in better prediction of what can be described as success
for a mobile app in today's time when there are hundreds of similar apps on the same platform.We have defined a success parameter
for an app based on the number of installs, distribution of ratings 4 and 5 relative to the overall number of ratings and installs to
rating ratio. Using features other than the ones used to create that parameter, we created a model that predicts it with 80.34%
accuracy.
REFERENCES
[1] Mobile App, https://en.wikipedia.org/wiki/Mobile_app [1]
[2] Venkata N Inukollu, Divya D Keshamoni, Taeghyun Kang and Manikanta Inukollu. “Factors Influencing Quality of Mobile Apps: Role of Mobile App
Development Life Cycle” in International Journal of Software Engineering & Applications (IJSEA), Vol.5, No.5, September 2014. [2]
[3] Abdul Mueez, Khushba Ahmed, Tuba Islam and Waqqas Iqbal, “Exploratory Data Analysis and Success Prediction of Google Play Store App,” B.Sc.
Engineering in CSE thesis, BRAC University, Dhaka - 1212, Bangladesh, December 2018. [3]
[4] Tuckerman, C.J., “Predicting Mobile Application Success,” Dec. 2014. [4]
[5] Federica Sarro, Mark Harman, Yue Jia, and Yuanyuan Zhang, “Customer Rating Reactions Can Be Predicted Purely Using App Feature”, CREST, Department
of Computer Science, University College London, London, UK.[5]
[6] (2002) The IEEE website. [Online]. Available: http://www.ieee.org/ [6]
[7] M. Shell. (2002) IEEEtran homepage on CTAN. [Online]. Available: http://www.ctan.org/tex-archive/macros/latex/contrib/supported/IEEEtran/ [7]
[8] FLEXChip Signal Processor (MC68175/D), Motorola, 1996. [8]
[9] “PDCA12-70 data sheet,” Opto Speed SA, Mezzovico, Switzerland. [9]
[10] A. Karnik, “Performance of TCP congestion control with rate feedback: TCP/ABR and rate adaptive TCP/IP,” M. Eng. thesis, Indian Institute of Science,
Bangalore, India, Jan. 1999. [10]
[11] J. Padhye, V. Firoiu, and D. Towsley, “A stochastic model of TCP Reno congestion avoidance and control,” Univ. of Massachusetts, Amherst, MA, CMPSCI
Tech. Rep. 99-02, 1999. [11]
[12] Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specification, IEEE Std. 802.11, 1997. [12]
[13] Liu, B. (2012). Sentiment analysis and opinion mining. Synthesis lectures on human language technologies.[13]
[14] Luiz, W., Viegas, F., Alencar, R., Mourão, F., Salles, T., Carvalho, D., Gonçalves, M. A., and Rocha, L. (2018). A feature-oriented sentiment rating for mobile
app reviews. In Proceedings of the 2018 World Wide Web Conference on World Wide Web, pages 1909–1918. International World Wide Web Conferences
Steering Committee. [14]
[15] Islam, M. R. (2014). Numeric rating of apps on google play store by sentiment analysis on user reviews. In 2014 International Conference on Electrical
Engineering and Information Communication Technology. [15]
[16] Grover, S. (2015). 3 apps that failed (and what they teach us about app marketing). [online] https://blog.placeit.net/apps-fail-teach-us-app-marketing/. [16]
[17] Fu, B., Lin, J., Li, L., Faloutsos, C., Hong, J., and Sadeh, N. (2013). Why people hate your app: Making sense of user feedback in a mobile app store. In
Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 1276–1284. ACM. [17]
[18] Prasad Patil (March 23, 2018). What is Exploratory Data Analysis? [Online]. In blog, Available: https://towardsdatascience.com/exploratory-data-analysis-
8fc1cb20fd15. [18]
[19] Lukas Frei (Apr 26, 2019). Speed Up Your Exploratory Data Analysis With Pandasa-Profiling [Online]. In blog, Available:
https://towardsdatascience.com/speed-up-your-exploratory-data-analysis-with-pandas-profiling-88b33dc53625. [19]