Designing and explaining the customer credit rating model using neural networks

Abstract

The consumer credit market in Iran has prospered with the establishment of private banks. The main activity in this market is granting consumer facilities to applicants, and this requires the validation of applicants for facilities in order to reduce credit risk. Nowadays, intelligent systems have found many applications in various banking and financial matters. Checking and approving credits is one of the applications of neural network. The present study is designed with the aim of providing a suitable model to investigate the credit behavior of customers of Mudarabah consumer facilities using neural networks for credit rating. Following this goal, first the important factors influencing the credit behavior of customers were identified and then the customers were divided into three categories: good accounts, bad accounts and past due. In the next step, neural network models after design; with training data; were trained and then tested with experimental data.

The obtained results show that the credit behavior of customers can be predicted using neural network ranking models.

Key words

Neural network; credit rating; Facilities

Introduction

The science of decision-making has always been with humans and has been greatly developed with the emergence of organizations, companies, and especially with rapid environmental changes. Many researchers have focused their efforts in this area to introduce more appropriate and accurate models to improve decision-making systems and make decision-makers more successful.[1]

In granting facilities, which is one of the main activities of banks and credit institutions, in order to make a correct decision, the degree of credit and ability to repay the principal and interest of the recipient should be determined in order to avoid the possibility of return of principal and interest. Granted facilities, i.e. credit risk, should be reduced. One of the ways to reduce this risk is to design a credit rating system for facility recipients, and the center of this system is the rating or credit evaluation model [2].

Using such a model, the rating or credit rating of the applicant is determined and based on that, a decision is made regarding whether or not to grant the facility. Currently, the use of intelligent systems for optimization and prediction is widely used as one of the advanced tools in various fields of science. Neural networks are used as an intelligent system in various financial fields, including credit approval.

In credit approval, customer credit evaluation is one of the most complex cases in financial activities[3].

It seems that the search for other practical relationships has lost its importance. What is important is to find out the movement and relationship of a set of variables with another set. For this, the artificial neural network model goes far beyond the brain, which cannot see everything together [4].

The credit evaluation of customers can be done by expert experts and evaluators, but this is often not cost-effective due to lack of time, high cost, lack of experts and the number of evaluation cases. By using information and communication technology, which has created a huge transformation in the banking system and while creating new opportunities, it has also brought new challenges, it is possible to design credit evaluation models that can separate good accounts (customers with good credit) and bad accounts (customers with bad accounts) using scientific methods instead of subjective judgments in a short time and at a reasonable cost.

1-1 Statement of the problem

Providing bank facilities is very important from an economic point of view. Because with a small increase in capital, it causes economic growth and development [5].  But in granting facilities, banks face a big risk called credit risk. This risk is the reason why banks face major financial crises. Credit risk can be considered as the possibility of non-repayment of the loan by the applicant [6].  which must be managed.Various methods can be used to manage credit risk. One of the methods is to design a credit rating system for facility recipients.

The purpose of this research is to identify customer behavior patterns in the credit market by designing and establishing a customer validation system and thus creating the possibility of predicting behavior. Customer credit evaluation is considered a very complex field of activities. The number of factors and the complexity of financial, economic and behavioral relationships make credit evaluation very difficult. On the other hand, the evaluation should often be done in a short time frame because the prolongation of the evaluation process will delay the operation and ultimately increase the costs. On the other hand, possible inaccuracy in the evaluation can lead to wrong decisions and ultimately huge losses. The time limit and the necessity of accuracy in the evaluation doubles the complexity of the issue[7].

Credit rating systems can be divided into three categories[8].

1- Judgmental systems

2 Ranking based on statistical techniques

3 systems Smart

Judging systems are very slow and expensive. Generally, when the number of requests is high, and the number of experts is low, these systems do not have the necessary efficiency, in the case of statistical methods, each of its techniques requires special assumptions. Obviously, with the absence or dimming of presuppositions, the accuracy and correctness of additions are doubted. Expert systems are of great help in solving problems when the decision rules are clear and the information is valid. But most of the loan granting institutions are not transparent and the information does not exist at all or part of the information is not correct, however, neural networks are a very suitable option. In Iran's credit market, one of the problems of granting facilities, the criteria for obtaining collateral or bringing cash on behalf of the applicants is the use of credits and facilities of the banking network. The analysis of the information shows that a higher percentage of the studied people consider the criteria for obtaining collateral and the inflexibility of the evaluation criteria to prevent the burning of the principal and profit of the facility as one of the problems of accessing the facilities and credits granted by the banking system. Also, a high percentage of the respondents stated that the long evaluation time is a problem[9].

According to the conditions of the credit market and taking into account the types of credit rating systems and the proper functioning of neural networks, the main problem of this research is to design a model with the help of neural network, which can be used to grant facilities in the bank with minimum errors and in the shortest possible time.

This model can have the necessary efficiency if it is able to give a suitable answer to the research questions.

1-2 research questions

1 Can the credit customers of the bank be ranked using the coordinates of the credit customers?

2 Can the credit customers of the bank be ranked using artificial neural networks 1-3 The importance and necessity of the research topic The necessity of the rating system will have the necessary importance when it has a suitable criterion [10] to evaluate customers before granting facilities, so that banking facilities are allocated to desirable customers using this system. From the point of view of the banking system [11], the ideal customer refers to customers who, while spending the received facilities in different economic sectors, can return the received facilities to the banking system on time. Failure to repay the facility on time indicates that the recipient of the facility has not been very successful in utilizing the facility received [12].

In simpler terms, the return from the use of the facility is less than its bank interest, therefore, it has faced problems when it comes to repayment. In this case, the bank has to borrow from other sources, including the central bank, at a rate higher than its deposit rate to compensate for the lack of liquidity.