Content uploaded by Pramila M. Chawan
Author content
All content in this area was uploaded by Pramila M. Chawan on Apr 27, 2017
Content may be subject to copyright.
ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 2, Issue 3, May 2013
503
Abstract— This paper describes about different data mining techniques used in financial data analysis. Financial data
analysis is used in many financial institutes for accurate analysis of consumer data to find defaulter and valid customer.
For this different data mining techniques can be used. The information thus obtained can be used for Decision making. In
this paper we study about loan default risk analysis, Type of scoring and different data mining techniques like Bayes
classification, Decision Tree, Boosting, Bagging, Random forest algorithm and other techniques.
Index Terms— Data mining, Bayes Classification, Decision Tree, Boosting, Bagging, Random forest algorithm
I. INTRODUCTION
In Banking Sectors and other such leading organization the accurate assessment of consumer is of uttermost
importance. Credit loans and finances have risk of being defaulted. These loans involve large amounts of capital
and their non-retrieval can lead to major loss for the financial institution. Therefore, the accurate assessment of the
risk involved is a crucial matter for banks and other such organizations. Not only is it important to minimize risk in
granting credit but also the errors in declining any valid customer. This is to save the banks from lawsuits.
Increasing the demand for consumer credit has led to the competition in credit industry. In India, there are an
increasing number of people opting for loans for houses and cars and there is also a large demand for credit cards.
Such credits can be defaulted upon and have a great impact on the economy of the country. Thus, assessing loan
default risk is important for our economy. Earlier Assessing of credit was usually done using statistical and
mathematical methods by analysts. Nowadays Data mining techniques have gained popularity over the years
because of their ability in discovering practical knowledge from the database and transforming them into useful
information.
II. LOAN DEFAULT RISK ANALYSIS
Loan default risk assessment is one of the crucial issues which financial institutions, particularly banks are faced
and determining the effective variables is one of the critical parts in this type of studies. In this Credit scoring is a
widely used technique that helps financial institutions evaluates the likelihood for a credit applicant to default on
the financial obligation and decide whether to grant credit or not. The precise judgment of the credit worthiness of
applicants allows financial institutions to increase the volume of granted credit while minimizing possible losses.
Credit scoring models are known as statistical models which have been widely used to predict the default risk of
individuals or companies. These are multivariate models which use the main economic and financial indicators of
a company or individuals' characteristic such as age, income and marital status as input, assign them a weight which
reflect its relative importance in predicting default. The result is an index of credit worthiness that is expressed as
a numerical score, which measures the borrower’s probability of default. The initial credit scoring models are
devised in the 1930s by authors such as Fisher and Durand. The goal of a credit scoring model is to classify credit
applicants into two classes: the “good credit" class that is liable to reimburse the financial obligation and the “bad
credit" class that should be denied credit due to the high probability of defaulting on the financial obligation. The
classification is contingent on characteristics of the borrower (such as age, education level, occupation, marital
status and income), the repayment performance on previous loans and the type of loan. These models are also
applicable to small businesses since these may be regarded as extensions of an individual costumer.
Type of scoring
Based on Paleologo, G., Elisseeff, A., and Antonini, G., research there are different kinds of scoring:
1) Application Scoring:
This kind of scoring consists of estimation credit worthiness of a new applicant who applies for credit. It estimates
financial risk with respect to social, demographic financial conditions of a new applicant to decide whether to
grand credit to them or not.
Study of Data Mining Techniques used for
Financial Data Analysis
Abhijit A. Sawant and P. M. Chawan
Department of Computer Technology, VJTI, Mumbai, INDIA
ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 2, Issue 3, May 2013
504
2) Behavioral Scoring:
It is similar to application scoring with difference that it involves existing customers, so the lender has some
evidences about borrower’s behavior to dynamic management of portfolio process.
3) Collection Scoring:
Collection scoring classifies customers into different groups according to their levels of insolvency. In the other
words, it separates the customers who need more decisive actions from those who do not require to be attended to
immediately. These models are used in order to management of delinquent customers from the first signs of
delinquency.
4) Fraud Detection:
It categorizes the applicants according to the probability that an applicant be guilty.
III. DATA MINING TECHNIQUES
Data mining a field at the intersection of computer science and statistics is the process that attempts to discover
patterns in large data sets. It utilizes methods at the intersection of artificial intelligence, machine learning,
statistics, and database systems. The overall goal of the data mining process is to extract information from a data set
and transform it into an understandable structure for further use. Aside from the raw analysis step, it involves
database and data management aspects, data preprocessing, model and inference considerations, interestingness
metrics, complexity considerations, post-processing of discovered structures, visualization, and online updating.
Mostly use Data mining techniques are as follows:
A. Bayes Classification
A Bayes classifier is a simple probabilistic classifier based on applying Bayes' theorem with strong independent
assumptions and is particularly suited when the dimensionality of the inputs is high. A naive Bayes classifier
assumes that the existence (or nonexistence) of a specific feature of a class is unrelated to the existence (or
nonexistence) of any other feature. Classification is a form of data analysis which can be used to extract models
describing important data classes. Classification predicts categorical labels (or discrete values).Data classification
is a two step process. In the first step, a model is built describing a predetermined set of data classes or concepts.
The model is constructed by analyzing database tuples described by attributes. Each tuple is assumed to belong to
a predefined class, as determined by one of the attributes, called the class label attribute. In the second step, the
model is used for classification. First, the predictive accuracy of the model is estimated. The accuracy of a model
on a given test set is the percentage of test set samples that are correctly classified by the model. . If the accuracy of
the model is considered acceptable, the model can be used to classify future data tuples or objects for which the
class label is not known.
Bayesian Algorithm:
1. Order the nodes according to their topological order.
2. Initiate importance function Pr0(X\E), the desired number of samples m, the updating interval l, and the score
arrays for every node.
3. k 0, TØ
4. for i 1 to m do
5. if (i mod l == 0) then
6. k k+1
7. update importance function Prk (X\E) based on T
end if
8. si generate a sample according to Prk (X\E)
9. T T U {si}
10. Calculate Score( si, Pr(X\E,e), Prk(X|E) and add it to the corresponding entry of every array according to the
instantiated states.
11. Normalize the score arrays for every node.
The major disadvantage of this model is that the predictive accuracy is highly correlated with this assumption.
An advantage of this method is that it requires a small amount of training data to estimate the parameters (means
and variances of the variables) necessary for classification.
B. Decision Tree
A classification (decision) tree is a tree-like graph of decisions and their possible consequences. Topmost node in
this tree is the root node on which a decision is supposed to be taken. In each inner node, it is done as a test on an
attribute or input variable. Specifically each branch of the tree is a classification question and the leaves of the tree
ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 2, Issue 3, May 2013
505
are partitions of the dataset with their classification. The processes in decision tree algorithms are very similar
when they build trees.
These algorithms look at all possible distinguishing questions that could possibly break up the original training
dataset into segments that are nearly homogeneous with respect to the different classes being predicted. Some
decision tree algorithms may use heuristics in order to pick the questions or even pick them at random. The
advantage of this method is that it is a white box model and so it is simple to understand and explain, but the
limitation of this model is that, it cannot be generalized into a designed structure for all contexts.
A decision tree is a mapping from observations about an item to conclusion about its target value as a predictive
model in data mining and machine learning. Generally, for such tree models, other descriptive names are
classification tree (discrete target) or regression tree (continuous target). In these tree structures, the leaf nodes
represent classifications, the inner nodes represent the current predictive attributes and branches represent
conjunctions of attributions that lead to the final classifications. The popular decision trees algorithms include ID3,
C4.5 which is an extension of ID3 algorithm and CART.
Fig 1- Decision Tree
C. Boosting
The concept of boosting applies to the area of predictive data mining, to generate multiple models or classifiers (for
prediction or classification), and to derive weights to combine the predictions from those models into a single
prediction or predicted classification.
A simple algorithm for boosting works like this: Start by applying some method to the learning data, where each
observation is assigned an equal weight. Compute the predicted classifications, and apply weights to the
observations in the learning sample that are inversely proportional to the accuracy of the classification. In other
words, assign greater weight to those observations that were difficult to classify (where the misclassification rate
was high), and lower weights to those that were easy to classify (where the misclassification rate was low).
Boosting will generate a sequence of classifiers, where each consecutive classifier in the sequence is an "expert" in
classifying observations that were not well classified by those preceding it. During deployment (for prediction or
classification of new cases), the predictions from the different classifiers can then be combined (e.g., via voting, or
some weighted voting procedure) to derive a single best prediction or classification.
The most popular boosting algorithm is AdaBoost. AdaBoost, short for Adaptive Boosting, is a machine
learning algorithm, formulated by Yoav Freund and Robert Schapire It is a meta-algorithm, and can be used in
conjunction with many other learning algorithms to improve their performance.
The Boosting Algorithm AdaBoost
Given: (x1, y1),…, (xm,ym); xi ϵ X, yi ϵ {-1,1}
Initialize weights D1 (i)=1/m
For t= 1, …, T:
1. (Call WeakLearn), which returns the weak classifier ht : X {-1,1} with minimum error w.r.t distribution Dt;
2. Choose αt ϵ R,
3. Update
ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 2, Issue 3, May 2013
506
Dt+1(i)=(Dt(i) exp (-αt yi ht(xi)))/Zt
Where Zt is a normalization factor Choosen so that Dt+1 is a distribution
Output the strong Classifier
H(x) =sign (∑T t=1 αt ht(x))
D. Bagging
Bagging (Bootstrap aggregating) was proposed by Leo Breiman in 1994 to improve the classification by
combining classifications of randomly generated training sets.
Bootstrap aggregating (bagging) is a machine learning ensemble meta-algorithm designed to improve the stability
and accuracy of machine learning algorithms used in statistical classification and regression. It also reduces
variance and helps to avoid over fitting. Although it is usually applied to decision tree methods, it can be used with
any type of method. Bagging is a special case of the model averaging approach.
Given a standard training set D of size n, bagging generates m new training sets Di, each of size n′ < n, by sampling
from D uniformly and with replacement. By sampling with replacement, some observations may be repeated in
each Di. If n′=n, then for large n the set Di is expected to have the fraction (1 - 1/e) of the unique examples of D, the
rest being duplicates. This kind of sample is known as a bootstrap sample. The m models are fitted using the above
m bootstrap samples and combined by averaging the output voting.
Bagging leads to "improvements for unstable procedures", which include, for example, neural nets, classification
and regression trees, and subset selection in linear regression? On the other hand, it can mildly degrade the
performance of stable methods such as K-nearest neighbours.
The Bagging Algorithm
Training phase
1. Initialize the parameter
D=Ø, the ensemble.
Lk the number of classifiers to train.
2. For k=1,…, L
Take a bootstrap sample Sk from Z.
Build a classifier Dk using Sk as the training set
Add the classifier to the current ensemble,
D=D U Dk.
3. Return D.
Classification phase
4. Run D1, …, Dk on the input x.
5. The clas with the maximum number of votes is chosen as the lable for x.
E. Random forest Algorithm
Random forests are an ensemble learning method for classification (and regression) that operate by constructing a
multitude of decision trees at training time and outputting the class that is the mode of the classes output by
individual trees. The algorithm for inducing a random forest was developed by Leo Breiman and Adele Cutler, and
"Random Forests" is their trademark. The term came from random decision forest which was first proposed by Tin
Kam Ho of Bell Labs in 1995. The method combines Breiman's "bagging" idea and the random selection of
features, introduced independently by Ho and Amit and Geman in order to construct a collection of decision trees
with controlled variation.The selection of a random subset of features is an example of the random subspace
method, which, in Ho's formulation, is a way to implement stochastic discrimination proposed by Eugene
Kleinberg.The introduction of random forests proper was first made in a paper by Leo Breiman. This paper
describes a method of building a forest of uncorrelated trees using a CART like procedure, combined with
randomized node optimization and baggingIt is better to think of random forests as a framework rather than as a
particular model. The framework consists of several interchangeable parts which can be mixed and matched to
create a large number of particular models, all built around the same central theme. Constructing a model in this
framework requires making several choices:
1. The shape of the decision to use in each node.
2. The type of predictor to use in each leaf.
3. The splitting objective to optimize in each node.
4. The method for injecting randomness into the trees.
F. Other Techniques
I] The Back Propagation Algorithm
ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 2, Issue 3, May 2013
507
The back propagation algorithm (Rumelhart and McClelland, 1986) is used in layered feed-forward ANNs. This
means that the artificial neurons are organized in layers, and send their signals “forward”, and then the errors are
propagated backwards. The network receives inputs by neurons in the input layer, and the output of the network is
given by the neurons on an output layer. There may be one or more intermediate hidden layers. The back
propagation algorithm uses supervised learning, which means that we provide the algorithm with examples of the
inputs and outputs we want the network to compute, and then the error (difference between actual and expected
results) is calculated. The idea of the back propagation algorithm is to reduce this error, until the ANN learns the
training data. The training begins with random weights, and the goal is to adjust them so that the error will be
minimal.
II] Genetic Algorithm:
Genetic Algorithm (GA) was developed by Holland in 1970. This incorporates Darwinian evolutionary theory with
sexual reproduction. GA is stochastic search algorithm modeled on the process of natural selection, which
underlines biological evolution. A has been successfully applied in many search, optimization, and machine
learning problems. GA process in an iteration manner by generating new populations of strings from old ones.
Every string is the encoded binary, real etc., version of a candidate solution. An evaluation function associates a
fitness measure to every string indicating its fitness for the problem. Standard GA apply genetic operators such
selection, crossover and mutation on an initially random population in order to compute a whole generation of new
strings.
Fig 2- Genetic Algorithm Flowchart
Selection deals with the probabilistic survival of the fittest, in those more fit chromosomes are chosen to survive.
Where fitness is a comparable measure of how well a chromosome solves the problem at hand.
Crossover takes individual chromosomes from P combines them to form new ones.
Mutation alters the new solutions so as to add stochasticity in the search for better solutions.
In general the main motivation for using GAs in the discovery of high-level prediction rules is that they perform a
global search and cope better with attribute interaction than the greedy rule induction algorithms often used in data
mining.
ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 2, Issue 3, May 2013
508
III] Particle Swamp Optimization:
The investigation and analysis on the biologic colony demonstrated that intelligence generated from complex
activities can provide efficient solutions for specific optimization problems. Inspired by the social behavior of
animals such as fish schooling and bird flocking, Kennedy and Eberhart designed the Particle Swarm Optimization
(PSO) in 1995. The basic PSO model consists of a swarm of particles moving in a d-dimensional search space. The
direction and distance of each particle in the hyper-dimensional space is determined by its fitness and velocity. In
general, the fitness is primarily related with the optimization objective and the velocity is updated according to a
sophisticated rule.
Artificial neural networks (ANNs) are thus, non-linear statistical modeling based on the function of the human
brain. They are powerful tools for unknown data relationship modeling. Artificial Neural Networks are able to
recognize the complex pattern between input and output variables then predict the outcome of new independent
input data.
IV] Support vector machine:
Support vector machine is a classifier technique. This method involves three elements. A score formula which is a
linear combination of features selected for the classification problem, an objective function which considers both
training and test samples to optimize the classification of new data, an optimizing algorithm for determining the
optimal parameters of training sample objective function.
The advantages of the method are that, in the nonparametric case, SVM requires no data structure assumptions
such as normal distribution and continuity. SVM can perform a nonlinear mapping from an original input space
into a high dimensional feature space and this method is capable of handling both continuous and categorical
predictions. The weaknesses of this method are that, it is difficult to interpret unless the features interpretable and
standard formulations do not contain specification of business constraints.
IV. CONCLUSION
The system using data mining for loan Default risk analysis enables the bank to reduce the manual errors involved
in the same. Decision trees are preferred by banks because they are a white box model. The discrimination made by
decision trees is obvious and the people can understand its working easily. This enables the banks and other
financial institutions to provide an account for accepting or rejecting an applicant. Boosting has already increased
the efficiency of decision trees. The assessment of risk will enable banks to increase profit and can result in
reduction of interest rate.
REFERENCES
[1] Jiawei Han, Micheline Kamber, “Data Mining Concepts and Technique”, 2nd edition
[2] Margaret H. Dunham,”Data Mining-Introductory and Advanced Topocs” Pearson Education, Sixtyh Impression, 2009.
[3] Abbas Keramati, Niloofar Yousefi, “A Proposed Classification of Data Mining Techniques in Credit Scoring",
International Conference on Industrial Engineering and Operations Management, 2011
[4] Boris Kovalerchuk, Evgenii Vityaev,” DATA MINING FOR FINANCIAL APPLICATIONS”, 2002
[5] Defu Zhang, Xiyue Zhou, Stephen C.H. Leung, JieminZheng, “Vertical bagging decision trees model for credit scoring”,
Expert Systems with Applications 37 (2010) 7838–7843.
[6] Raymond Anderson, “Credit risk assessment: Enterprise-credit frameworks”.
[7] Girisha Kumar Jha, ”Artificial Neural Networks”.
[8] Hossein Falaki ,”AdaBoost Algorithm”.
[9] Venkatesh Ganti, Johannes Gehrke, Raghu Ramakrishnan, “Mining Very Large Database” IEEE,1999.
[10] http://en.wikipedia.org.
AUTHOR BIOGRAPHY
Abhijit A. Sawant is currently pursuing M. Tech in Computer engineering second year from “Veermata Jijabai
Technological Institute (V.J.T.I), Matunga, Mumbai (INDIA)”. He has received his Bachelors’ Degree in Information
technology from “Padmabhushan Vasantdada Patil Pratishan’s College of Engineering (P.V.P.P.C.O.E), Sion, Mumbai
(INDIA)” in 2010. He has published 2 papers in International Journal till date. His areas of interest are Software
Engineering, Database, Data ware house, Data mining.
ISSN: 2319-5967
ISO 9001:2008 Certified
International Journal of Engineering Science and Innovative Technology (IJESIT)
Volume 2, Issue 3, May 2013
509
Pramila M. Chawan is currently working as an Associate Professor in the Computer Technology Department of
“Veermata Jijabai Technological Institute (V.J.T.I.), Matunga, and Mumbai (INDIA)”. She received her Bachelors’
Degree in Computer Engineering from V.J.T.I., Mumbai University (INDIA) in 1991 & Masters’ Degree in Computer
Engineering from V.J.T.I., Mumbai University (INDIA) in 1997.She has an academic experience of 20 years. She has
taught Computer related subjects at both Undergraduate & Post Graduate levels. Her areas of interest are Software
Engineering, Software Project Management, Management Information Systems, Advanced Computer Architecture &
Operating Systems. She has published 12 papers in National Conferences and 7 papers in International Conferences &
Symposiums. She also has 40 International Journal publications to her credit. She has guides 35 M. Tech. projects &
85 B. Tech. projects.
