International Journal of Engineering and Technical Research (IJETR)

ISSN: 2321-0869 (O) 2454-4698 (P), Volume-3, Issue-7, July 2015

An Overview of Data Mining In Medical Field

Reema Arora , Sandeep Jaglan

objects in database, if the database is a reliable mirror of the

Abstract Data mining is a upward area of research that real world registered by the database.
intersects with many disciplines such as Artificial Intelligence
(AI), databases, parallel computing , statistics, visualization, and II. DATA MINING PROCESS
high-performance. The target of data mining is to turn data that
are specifics, records, or text which can be processed by a The data mining process may be intricate and can be divided
computer into knowledge. Nowadays, the reliance of health care into the following steps:
on data is mounting. Therefore, this paper aims to allow the 1. Domain analysis and data understanding
readers to comprehend about data mining and its importance in
2. Data selection
medical systems. The ambition of studying data mining
techniques for the diagnosis and prognosis of various diseases is 3. Data analysis and pre-processing
to identify the well-performing data mining algorithms used on 4. Data reduction and transformation
medical databases. The subsequent algorithms have been
acknowledged: Decision Trees, Artificial neural networks and 5. Important attributes selection
their Multilayer Perception model, Nave Bayes. Analyses show 6. Reduction of the number of dimensions
that it is very thorny to name a single data mining algorithm as
the most suitable for the diagnosis or prognosis of diseases. At 7. Normalization
times some algorithms execute superior than others, but there 8. Aggregation
are cases when a grouping of the best properties of some of the 9. Selection of data mining method
algorithms mentioned above collectively results more valuable. 10. Data mining process
11. Visualization
Index Terms Nave Bayes, Multilayer Perception, C4.5, 12. Evaluation
medical data mining, medical decision support. 13. Knowledge utilization and evaluation of the
results to an appropriate target.
The masters thesis focuses on the ninth- the selection of data
I. INTRODUCTION mining methods. This selection is based on an in-depth
Data mining: - The term data mining refers to the finding of analysis of the methods performance measured with the use
relevant and helpful information from databases. Data mining of several metrics, like ROC curves, false or true, optimistic
and knowledge discovery in the database is a new or pessimistic rates.
interdisciplinary field, integrating facts from statistics,
machine learning, database and parallel computing III. DATA MINING TECHNIQUES
.Researchers have defined the term data mining in many ways: There are several major data mining techniques being
1. Data mining and knowledge discovery in databases, as it is developing and using in data mining projects including
also well-known is the non-trivial pulling out implicit, association, classification, sequential patterns ,clustering
previously unfamiliar and potentially valuable information and decision tree. We will look at those data mining
from the data. This encompasses a numeral of methodological technique in brief as follows:
approaches, such as clustering, data summarization,
classification, discovering dependency networks, analysing Classification:
changes, and detecting anomalies. It is the demonstration of data in given classes which is called
supervised classification, it uses given class labels to order the
Data retrieval in its common sense in database attempts to objects in the data collection. Classification consider as an
retrieve data that is stored visibly in database and present it to imperative task of data mining. Using this approach data must
the user in a way that user can understand it. It does not be defined as class label (target attribute). In binary
attempt to extract implicit information classification, the target attribute has only two possible
Data mining is the seek for the relationships and global values: for example, high or low. Multiclass targets have more
patterns that exist in the large databases but are hidden among than two values: for example, low, medium, high, or
unknown. Classification can applied into Business modeling,
enormous amount of data, such as the bond between patient marketing, credit analysis, biomedical and drug response
information and their medical diagnosis. This relationship modeling.
represents important knowledge about the database and the
Clustering

Clustering is the depiction of data in classes. However, not

like classification, in clustering class labels are mysterious
Reema Arora, Computer Science and Engineering, N.C. College of
Engineering, Israna Panipat, India and it is up to the clustering algorithm to determine acceptable
Sandeep Jaglan, Computer Science and Engineering, N.C. College of classes. This is called unsupervised classification. Clustering
Engineering, Israna Panipat, India

198 www.erpublication.org
 An Overview of Data Mining In Medical Field

is a collection of data objects, similar data are taking in the deeply a tree can grow. The C4.5 algorithm is capable of
same cluster, dissimilar data are taking in different clusters. managing continuous attributes, which are vital in case of
medical data (e.g. blood pressure, temperature, etc.). Other
Association: very familiar aspect missing values was also taken into
consideration in C4.5. Moreover the algorithm handles
Association analysis is the unearthing of association rules.
attributes with differing costs.
It depends on the occurrence of transactional data occur
The utility of C4.5 algorithm was widely proven in medicine .
together in database, also depends on a threshold called
This algorithm suits medical data because it copes with
support, and identifies the frequent item sets. Association data
missing values. What is more the algorithm handles
mining designed to find association between attributes,
continuous data which are common among medical
generate rules from data sets. The association rule mining role
symptoms. The efficiency of C4.5 was shown e.g. in breast
is to arrive at all rules having support minsup (minimum
cancer and prostate cancer classification to generate a
support) threshold and confidence minconf (minimum
decision tree and rules which may be helpful in medical
confidence) threshold.
diagnosing process.
Sequential Patterns
A Nave Bayes
Sequential patterns analysis is another data mining A nave bayes classifier assumes that the occurrence (or
technique that seeks to discover or classify similar patterns, absence) of a particular feature of a class is unrelated to the
regular events or trends in transaction data over a business presence (or absence) of any other feature. For illustration, a
period. fruit may be assumed to be a tomato if it is red in colour, round
in shape, and about 2.5" in diameter. This classifier takes all
Decision trees: these description to contribute separately to the probability
that this fruit is a tomato, whether or not they're in fact related
Decision tree is one of the most used data mining techniques to each other or to the existence of the other features. The
because its model is easy to comprehend by users. In decision Bayes theorem is as follows: Let X={x1, x2,.....,xn} be a set
tree technique, the root of the decision tree is a simple of n attributes. In Bayesian, X is assumed as proof and H be
question or condition that has numerous answers. Each some hypothesis means the data of X belongs to precise class
answer then leads to a set of questions or conditions that help C. To determine P (H|X), the probability that the hypothesis H
us determine the data so that we can make the final decision holds specified facts i.e. data sample X. According to Bayes
based on it. theorem the P (H|X) is expressed as P (H|X) = P (X| H) P (H)
/ P (X) As Nave Bayes classifiers depends on the precise
nature of the probability model , so it can be trained very
efficiently in a supervised learning setting. Here independent
variables are considered for the principle of prediction or
occurrence of the event. It has been shown that Nave Bayes
classifiers often work much better in many complex real
world situations. An assistance of the Nave Bayes classifier is
that it requires a small amount of training data to estimate the
parameters (means and variances of the variables) necessary
for classification.

V. NEURAL NETWORKS

Artificial neural networks (ANNs) are based on study of a

human brain. Human brain is a very complicated web of
IV. DATA MINING ALGORITHMS APPLIED IN MEDICAL neurons. Analogically, the ANN is an interconnected set of
SYSTEM neurons simple units which are of three types: input,
concealed and output ones. The path of the network
Data mining also identified as Knowledge Discovery in
symbolizes the data flow during the course of prophecy. The
databases is very often utilized in the field of medicine. The
attributes that are passed as input to the net form a first layer.
practice of supporting medical diagnoses by robotically
In medical diagnosis the input to the artificial neural network
searching for valuable patterns undergoes noticeable
are the patients symptoms. The set S and an output is the
improvements in terms of precision and response time. This
analysis: the set D. The concealed neuron practices the
paper shortly describes the most common data mining
outcomes of earlier layers. The connections between neurons
algorithms and explains the use cases of each of them. The
have weights assigned to them. Their values are accustomed
usefulness of the following methods was verified by medical
with the use of appropriate algorithms, like back propagation.
personnel and confirmed by independent experts. The
It uses a gradient descent to tune networks parameters to best
algorithms are as follows:
fit the values of the output. The veiled layers add nonlinear
The C4.5 algorithm
characteristics to the network. The output layer may have
more than one output node (prediction of different diseases).
C4.5 is an extension of the well-known ID3 algorithm. The
One of the main advantages of Artificial Neural Networks is
extension includes avoiding data over fitting by shaping how

199 www.erpublication.org
 International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869 (O) 2454-4698 (P), Volume-3, Issue-7, July 2015
their high performance. The drawback of this method is its Knowledge-Based MDSS Most MDSS are divided into three
complexity and difficulty in understanding the predictions. parts, the knowledge base, inference engine and mechanism to
communicate. The knowledge base possessed the IF-THEN
rules. The inference engine gather the rules from the
knowledge base with the patients data. The communication
mechanism will permit the system to show the results to the
user as well as have input into the system. Features of a
non-Knowledge-Based MDSS Two types of
non-knowledge-based systems are neural networks and
genetic algorithm. Neural networks use nodes and weighted
connections between them to analyze the patterns found in the
patient data to develop the associations between the
symptoms and a diagnosis. Genetic Algorithms are based on
basic evolutionary processes using directed selection to
Multilayer perceptions (MLPs) achieve optimal MDSS results. The MDSS features
associated with success include the following:
Multilayer perceptions are feedforward neural networks It is incorporated into the health care workflow rather than
skilled with the paradigm backpropagation algorithm. They as a separate log-in or screen.
are called as supervised networks because they require a
desire reply to be skilled. They gain knowledge of how to It is electronic unlike paper-based templates.
transform input data into a desired response,, therefore they
It gives decision support at the time and location of care
are widely used for pattern classification. With single or two
rather than prior to or after the patient encounter.
hidden layers, they can fairly accurate virtually any
input-output map. They have been revealed to approximate It gives(active voice) recommendations for care, not just
the performance of optimal statistical classifiers in difficult assessments.
problems. Most neural network applications involve MLPs.

VI. MEDICAL DECISION SUPPORT SYSTEM IN DATA MINING VII. CHARACTERISTICS OF MEDICAL DECISION SUPPORT
Data mining applications are presently being applied to two SYSTEMS
main branches in health care and medicine: Medical decision The Medical DSSs are the type of computer programs that
support system, and policy planning/decision making. A. help out physicians and medical staff in Medical decision
Medical decision support system MDSS is an interactive making tasks. Most of the Medical decision support systems
Decision support system (DSS) Computer Software, which is (MDSSs) are equipped with diagnostic assistance module,
intended to lend a hand to physicians and other health therapy critiquing and planning module, medications
professionals with judgment making tasks, such as prescribing module, information retrieval subsystem (for
determining diagnosis of patient data. . The main purpose of instance formulating accurate clinical questions) and image
modern MDSS is to help clinicians at the point of care. It recognition and interpretation section (X-rays, CT, MRI
means, a clinician would interact with a MDSS to help scans) Interesting examples of MDSSs are machine learning
determine diagnosis, analysis, etc. of patient data. It is a systems which are able to create new healthcare knowledge.
decision-support system program that offers employees in By analyzing healthcare cases a Medical Decision Support
detail, purpose, custom-made, and current information on all System can produce a detailed description of input features
healthcare conditions. Employees receive the information, with a unique characteristic of healthcare conditions. It
implements and support they need from incorporated web, supports may be priceless in looking for changes in patients
phone, and print based materials. This helps employees health condition. These systems may improve patients safety
formulate more informed healthcare decisions while working by reducing errors in diagnosing. They may also get enhanced
with their own physician. An example of how a MDSS might medications and test ordering. Furthermore, the quality of
be helpful in medicinal gather from the subset of Medical care gets better due to the lengthening of the time clinicians
Decision Support System and Diagnosis Decision Support spend with a patient. It may be an effect of application of
Systems. A DDSS would obtain the patients data and proper guidelines, up-to date healthcare evidence and
recommend a set of correct diagnoses. The doctor then takes improved documentation. Moreover, the efficiency of the
the output of the DDSS and point out which are relevant and health care delivery is improved by reducing costs through
which are not. Another important classification of a MDSS is faster order processing or eliminated duplication of tests.
based on the timing of its use. Doctors apply these systems at Examples of Medical Decision Support Systems
point of care to aid them as they are handling a patient, with There exist several Medical Decision Support Systems
the time of use as either pre-diagnoses, during diagnoses, or (MDSSs). They help in early detection of diseases. In this
post diagnoses. Pre-diagnoses MDSS systems are used to help survey a few of the most significant systems are accessible.
the physician prepare the diagnoses. MDSS helpful during They are utilized in hospitals. To provide you the idea of
diagnoses in reviewing and filtering the physicians Medical Decision Support Systems three sample ones are
preliminary diagnostic choices to improve their final results. described: HELP, DX plain and ERA.
And post-diagnoses MDSS systems are used to mine data to HELP
derive connections between patients and their past medical One of the most accepted and advanced Medical Decision
history and to predict future events. Features of a Support System is called HELP. It helps the clinicians in

200 www.erpublication.org
 An Overview of Data Mining In Medical Field

interpreting healthcare information, diagnosing the disease of IX. CONCLUSION

patients, maintaining healthcare protocols and other tasks .In Nowadays, massive amount of data is gathered in medical
2003 a new version was released called HELP II. It is databases Such databases may contain important information
equipped with a knowledge database which stores about contained in nontrivial dependencies among symptoms and
32000 emergency cases and a Medical decision support diagnoses. By using medical systems the process of revealing
engine. This system contains two assistants called antibiotic such relationships in historical data is become much easier to
assistant and pneumonia diagnostic assistant. The idea of the conduct. This knowledge can be valuable in diagnosis of
former is to find the pathogens causing the infection and to future cases.
suggest the cheapest therapy for patients with e.g. allergies or The main purpose of the research was to recognize the most
renal functions. DX plain It is a Medical Decision Support common data mining algorithms, implemented in modern
System (MDSS) available through the World Wide Web that Medical Decision Support Systems, and evaluate their
assists clinicians by generating stratified diagnoses based on performance on several medical datasets. Three algorithms
user input of patient signs and symptoms, laboratory results, were chosen: C4.5, Multilayer Perception and Nave Bayes.
and other healthcare findings Each healthcare finding entered For the evaluation five UCI databases were used: heart
into DX plain is assessed by determining the importance of disease, dermatology diseases, hepatitis, breast cancer and
the finding and how strongly the finding supports a given diabetes datasets. Several performance metrics were utilized:
diagnosis for each disease in the knowledge base.By Using percent of accurate classifications, True or False optimistic
this criterion, DX plain produces ranked differential rates, AUC, Precision, Recall, F-measure and a set of errors.
diagnoses with the most likely diseases compliant the lowest The underlying cause for such a research was the fact that no
rank. ERA (Early Referrals Application) The Early Referrals work was found which would analyze these three algorithms
Application (ERA) is one of the newest and most promising under identical conditions.
Medical Decision Support Systems. This clarification is
devoted to detection of different types of cancers in their early REFERENCES
stage.
[1] Aftarczuk K., Kozierkiewicz A., The method of supporting medical
VIII. DATA MINING CHALLENGES IN HEALTHCARE diagnosis based on consensus theory. Report of Institute of
One of the most significant challenges of the data mining in Information Science & Engineering, University of Technology.
Wroclaw, 2006 Series
healthcare is to obtain the quality and relevant medical data. It [2] Herron P., Machine Learning for Medical Decision Support:
is not easy to acquire the exact and complete healthcare data. Evaluating Diagnostic Performance of Machine Learning
Health data is complex and heterogeneous in nature because it Classification Algorithms, INLS 110, Data Mining, 2004.
is collected from various sources such as from the medical [3] W. Lord and D. Wiggins, "Medical Decision Support Systems
Advances in Health care Technology Care Shaping the Future of
reports of laboratory, from the conversation with patient or
Medical." vol. 6, G. Spekowius and T. Wendler, Eds., ed: Springer
from the review of medical doctor. For healthcare provider, it Netherlands, 2006, pp. 403-419
is crucial to maintain the quality of data because this data is [4] J. Han and M. Kamber, Data Mining, concepts and techniques, 1st ed.:
useful to offer cost effective healthcare treatments to the Academic Press, 2001.
patients. Health Care Financing Administration helps in [5] M. Kantardzic, Data mining: concepts, models, methods, and
algorithms: Wiley-IEEE Press, 2003
maintaining the minimum data set (MDS) which is recorded [6] Chen, M.S., Han, J. and Yu, P. (1996), ``Data mining: an overview
by every hospital. In MDS there are 200 questions which are from a database perspective.
answered by the patients at register time. But this practice is [7] Fayyad, U., Piatetsky-Shapiro, G. and Smyth, P. (1996), ``From data
difficult and patients face problems to respond the entire mining to knowledge discovery: an overview, in Fayyad, U.,
Piatestsky-Shapiro, G., Smyth, P. and Uthurusamy, R. (Eds),
questions. Due to this MDS face some difficulties such as Advances in Knowledge Discovery and Data Mining, MIT Press,
missing information and incorrect entries. Without quality Cambridge, MA.
data there is no useful results. For successful data mining, [8] Amandeep Kaur Mann, Navneet Kaur, Survey Paper on Clustering
complication in medical data is one the significant hurdle for Techniques, IJSETR, 2278 779. [2] Pavel Berkhin, A Survey of
Clustering Data Mining Techniques, pp.25-71, 2002
analyzing medical data. So, it is essential to maintain the [9] Oded Maimon, Lior Rokach, Data Mining AND Knowledge
quality and accuracy data for data mining to making effective Discovery Handbook, Springer Science + Business Media.Inc,
decision. Another difficulty with healthcare data is data pp.321-352, 2005.
sharing. Healthcare organizations are unwilling to share their [10] Jenn-Lung Su, Guo-Zhen Wu, I-Pin Chao (2001). The Approach of
Data Mining Methods For Medical Database. IEEE. p1-3.
data due to privacy concern. Most of the patients do not want [11] Koh, Hian Chye, and Gerald Tan. "Data mining applications
to disclose their health data. Therefore, the Health in healthcare."Journal of Healthcare Information Management Vol
Maintenance Organization and Health insurance 19.2 (2011): 65.
Organization are not distributing their data for preserving the
privacy of patient. This poses hurdle in the fraud detection
studies in health insurance. The startup cost of data warehouse
is very high. Before applying data mining techniques in
healthcare data it is important to accumulate and record the
data from different sources into a central data warehouse
which is a costly and time overriding process. flawed data
warehouse design does not supply effective data mining.

201 www.erpublication.org