International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169

Volume: 5 Issue: 7 293 296

_______________________________________________________________________________________________
Prediction of Default Customer in Banking Sector using Artificial Neural
Network

W.S.Kanmani B.Jayapradha
Research Scholar PG and Research Department of Computer Assistant Professor PG and Research Department of Computer
Science, Dr.Ambethkar Government Arts College, Science, Dr.Ambedar Government Arts College,
Chennai 600 039, India. Chennai 600 039, India.
e-mail: [email protected] e-mail:[email protected]

Abstract The aim of this article is to present perdition and risk accuracy analysis of default customer in the banking sector. The neural network
is a learning model inspired by biological neuron it is used to estimate and predict that can depend on a large number of inputs. The bank
customer dataset from UCI repository, used for data analysis method to extract informative data set from a large volume of the dataset. This
dataset is used in the neural network for training data and testing data. In a training of data, the data set is iterated till the desired output. This
training data is cross check with test data. This paper focuses on predicting default customer by using deep learning neural network (DNN)
algorithm.
Keywords-Neural Network, Deep learning neural network(DNN), Rapid Miner, DefaultCustomer

__________________________________________________*****_________________________________________________
I. INTRODUCTION II. LITERATURE REVIEW
In recent research, works evolve around with neural Burez and Van den Poel indicate that there are two
network. So the neural network has enormous developing in types of targeted approaches to managing customer churn:
every field because Neural Network learns to do the task by reactive and proactive. When a company adopts a reactive
observational data without task specification programs. It approach, it waits until customers ask the company to cancel
typically has a number of neuron and connections those are their service relationship. On the other hand, when a
divided by layers. Deep learning Neural Network (DNN) is a company adopts a proactive approach, it tries to identify
powerful technique for a neural network and it provides the customers who are likely to default before they do so.
best solution for many problems. At present in the field of Targeted proactive programs have potential advantages of
business, customer relation managements first step is to having lower incentive costs [4]. In the financial area, the
create a customer retention analysis model. Most of the multilayer perceptron (MLP) network trained by the back-
businesses fields hope to figure out the real cause of lost a propagation learning algorithm is the most used technique for
customer, or even to be told that they are about to lose the financial decision-making problems [10]. In existing system
customer by some clues before it occurs, and then they can propose a method to solve such a problem, and according to
propose or make some new sales strategies against the loss in the method, to find out behavior patterns of losing customers
advance [7].Banks have realized that customer relations are a or clues before they stop using some products through
very important factor for their success. Constants increase in Mining Sequential Patterns [7]. In existing paper Multilayer
customers, increase the volume of data stored in banking perceptron used to find default customer. It has two different
database. Those data may not provide sufficient information approaches based on neural networks for identifying
so the bank faces a challenge to find the default customer and important variables. The first is based on error change and
how to retain most profitable customers. Data analysis the second is based on weights contribution in the network
techniques are resourceful for extracting the needed [5]. Another existing work proposed back propagation feed
information [3]. This paper focuses on predicting the default forward to predict customers. It also provides a solution for
customer from customer data set in banking sector using overtraining problem that occurs in the neural networks[6].
Deep Learning Neural Network with help of Rapid Miner
studio software package which provides integrated solution III. RELATED WORK
for neural network and data analysis. The hypothesis was that According to research literature about default customer,
customer who often uses bank transaction are more loyal and most of the related work focuses on using only one data
the bank should focus on a customer who uses the least mining method such as classification or clustering to mine
transaction and make them turn to be a profitable customer. the customer retention data. From the review of the
literature, we conclude that neural network can predict
default customer in different domain like cellular network

293
IJRITCC | July 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
 International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 7 293 296
_______________________________________________________________________________________________
[6], pay TV [4], telecom [5], credit risk analysis [8], and In network training phase we are selecting number
bankrupt [9]. hidden layers and number of training cycles. The software
This paper proposed deep neural network based approach package offers the best model which we can edit. Figure.1
to predict the default customer in banking and following shows our work uses three hidden layers with 9(8 input+1
hypotheses question are consider for predictions thresholds) input neuron and two output results.
How much transaction does the customer perform per
month?
What is the occupation of the customer?
What is a profile for the high-risk borrower?
How much Balance maintain each month?
What is the private status of a customer?
IV. METHODOLOGY
a. Rapid Miner
Rapid Miner is a commercial data science software
platform developed by the company of the same name
that provides an integrated environment for machine
learning, deep learning, text mining, and predictive
analytics. It is used for business and commercial Figure 1. Network Topology in Rapid Miner
applications as well as for research, education, training, In figure 2 shows the deep learning neural network training
rapid prototyping, and application development and and testing phase with parameter setting.
supports all steps of the machine learning process
including data preparation, results in visualization,
validation, and optimization.

b. Data Analysis
The used data set contain a large number of data
collection which consists of noisy, irrelevant, missing
values. We extract need data set by using one of data
analysis method, this data set which can apply in neural
network training. The extracted data set is 3846 based on
the client age, job, day and month which most frequent.
c. Deep Learning Neural Network
Deep learning is part of machine learning, is an
Figure 2. Deep learning in rapid miner
application to learn tasks of artificial neural network
(ANN). It allows computational models that are After training complete we get the results that shown in
composed of multiple processing layers to learn Table 1. With overall Accuracy: 78.18%
representations of data with multiple levels of
Table 1.Performance Accuracy Table
abstraction. These methods have dramatically improved
in many domains like speech recognition, visual object Class
True no True yes
recognition, object detection and other domains [12]. It Precision
is very best for supervised learning. DNN has common
Pred.No 2775 305 90.10%
issues are overfitting and computation time this is due to
naively training of data [13]. These can be overcome Pred.Yes 534 231 30.20%
with well train data sets.
Class recall 83.86% 43.10%
V. RESULT ANALYSIS
The paper used rapid miner software package which Table 1. shows the Overall accuracy of actual prediction
supports neural network to detect default customer. After achieves by with three hidden layer models with six neurons
selecting a data set, the software goes to data analysis phase, is roughly trained with the result of 78.18%. the dataset is
divided into 10 sets of the model and one testing model.
we are extracting need data and defining target output field
each learning is iterated 10 times with one testing models on
which applied in a neural network for prediction. each set. The prediction made rough of 1000 example after
being trained on nearly about 50 is nearly 78%. +/-1.09% is
standard deviation calculated from 10 individual model
294
IJRITCC | July 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
 International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 7 293 296
_______________________________________________________________________________________________
accuracy errors on 10 models which are trained. The class VI. DISCUSSION
precision shows the work of correct prediction for each This work concludes that there is a problematic group of
label. Figure 3 shows the chart result. people (self-employed, entrepreneur) with less transaction in
a month also have credit balances, who in the future can
become very important and very valuable clients. The bank
3000 should focus on offering services to these clients. For
2500 example, Bank could introduce new scheme tailored to self-
employed and entrepreneur needs such as car loan with
prediction

2000 insurance, favorable interest rates, promotional use of

pred.No
1500 internet banking, etc. Also, the DNN topology of the neural
pred.Yes network gives the better results.
1000
Class recall
500
VII. CONCLUSION AND FUTURE ENHANCEMENT
0
Building an effective default customer prediction model
True No True Yes class prection using various techniques has become a significant topic for
actual business and academics in recent years. Default prediction
and management is a curial task in the banking sector. In
order to compete in the financial field, the bank has to be
Figure 3. Prediction Results
able to predict possible defaults and take proactive actions to
Confusion matrix calculation is shown in Table 2 with
retain the valuable loyal customer. The results show that The
respected bar chart result shown in Figure 3. customer who uses bank often are more loyal and who uses
Table 2. Confusion matrix table bank very less turns to be defaulters. It also shows the
accuracy rate of the defaulter. In future work, several issues
can be considered and it can be implemented along with
Predict. No Predict. Yes Error neural network several other popular prediction techniques
can be applied in combination such as vector machine,
Actual. No 2846 463 0.1399 machine learning to develop hybrid models.

Actual. References
284 252 0.5299 [1] AlisaBilalZori Interdisciplinary Description of Complex
Yes
Systems Predicting Customer Churn In Banking Industry
Using Neural Networks vol 14(2), pages116-124, 10
Totals 3130 715 0.1943
March 2016.
[2] Sumathi, S. and Sivanandam, S.N.: Introduction to Data
In Table 2. shows the confusion matrix for the actual and Mining Principles. Studies in Computational Intelligence
prediction with error mean values for each prediction. The 29(3), 1-20 2013,
total outcome of 3130 data the error is 0.19% its shows the [3] Domingo, R.: Applying data mining to banking.
efficient in reducing error for better prediction results Http://www.rtdonline.com, accessed 18th November 2015,
[4] Burez, J., & Van den Poel, D. (2007). CRM at a pay-TV
company: Using analytical models to reduce customer
attrition by targeted marketing for subscription services.
Expert Systems with Applications, 32, 277288.
3000
[5] Omar Adwan, Hossam Faris, Khalid Jaradat, Osama
2500 Harfoushi and Nazeeh Ghatasheh. Predicting Customer
Churn in Telecom Industry using MLP Neural Networks:
2000
Prediction

Modeling and Analysis. Life Sci J 2014;11(3):75-81

No [6] Chiang, D., Wang, Y., Lee, S., & Lin, C. (2003). Goal-
1500
oriented sequential pattern for network banking churn
1000 Yes analysis. Expert Systems with Applications, 25(3),293302.
[7] Nazari, M. and Alidadi, M.: Measuring credit risk of bank
500 customers using artificial neural network. Journal of
0 Management Research 5(2), 17-27, 2013,
[8] Salvatore Madonna, Prof,Greta Cestari, PhD THE
No Yes Error ACCURACY OF BANKRUPTCY PREDICTION
Actual MODELS: A COMPARATIVE ANALYSIS OF
MULTIVARIATE DISCRIMINANT MODELS IN THE
ITALIAN CONTEXT European Scientific Journal
Figure 4. Confusion Matrix Chart With Error Mean

295
IJRITCC | July 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 7 293 296
_______________________________________________________________________________________________
December 2015 edition vol.11, No.34 ISSN: 1857 7881 Journal of Management Research, ISSN 1941-899X,2013,
(Print) e - ISSN 1857- 7431. Vol. 5, No. 2.
[9] Tsai, C.F. and Wu, J.W.: Using neural network ensembles [14] Bar, M.V.: The Computational Intelligence Techniques For
for bankruptcy prediction and credit scoring. Expert Predictions-Artificial Neural Networks. Annals of
Systems with Applications 34(4), 2639-2649, 2008. Computational Economics 2(42), 184-190, 2014
[10] Wilson, R.L. and Sharda, R.: Bankruptcy prediction using [15] Anuj Sharma, Dr.Prabin Kumar Panigrahi. A Neural
neural networks. Decision Support Systems 11(5), 545-557, Network Based Approach for Predicting Customer Churn
1994. in Cellular Network Services.International Journal of
[11] Deep Learning. Yann LeCun, Yoshua Bengio & Geoffrey Computer Applications (0975 8887)Volume 27 No.11,
Hinton. Nature 521, 436444 (28 May 2015) August 2011M. Young, The Technical Writers Handbook.
[12] Ivakhnenko, Alexey (1971). "Polynomial theory of Mill Valley, CA: University Science, 19.
complex systems". IEEE Transactions on Systems, Man
and Cybernetics (4): 364378.
[13] Mohsen Nazari , Mojtaba Alidadi Measuring Credit Risk
of Bank Customers Using Artificial Neural Network

296
IJRITCC | July 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________