Differentiating between good credits and bad credits using neuro-fuzzy systems

doi:10.1016/S0377-2217(01)00052-2

European Journal of Operational Research

Volume 136, Issue 1, 1 January 2002, Pages 190-211

https://doi.org/10.1016/S0377-2217(01)00052-2 Get rights and content

Abstract

To evaluate consumer loan applications, loan officers use many techniques such as judgmental systems, statistical models, or simply intuitive experience. In recent years, fuzzy systems and neural networks have attracted the growing interest of researchers and practitioners. This study compares the performance of artificial neuro-fuzzy inference systems (ANFIS) and multiple discriminant analysis models to screen potential defaulters on consumer loans. Using a modeling sample and a test sample, we find that the neuro-fuzzy system performs better than the multiple discriminant analysis approach to identify bad credit applications. Further, neuro-fuzzy systems have many advantages over traditional computational methods. Neuro-fuzzy system models are flexible, more tolerant of imprecise data, and can model non-linear functions of arbitrary complexity.

Introduction

Between 1985 and 1996, the number of personal bankruptcy cases filed in the US rose from 341,000 to 1.1 million, and the rate of bankruptcy per 100,000 adults increased from 203 to 596 (Nelson, 1999). At the same time, competition in the consumer credit market has become intense. Consequently, financial institutions are faced with the dilemma of trying to increase loan volume without unduly increasing their exposure to default. Therefore, to screen loan applications, new techniques such as neuro-fuzzy models (compared to the traditional statistical models) should be explored to help predict potential default loans more reliably.

The loan officers of financial institutions use many techniques such as statistical models, credit scoring models, or rules that have been developed through experience to evaluate consumer loan applications. The use of a technique depends on the complexity of the institution, and the size and the type of the loan. Analytical models, such as empirically derived credit scoring systems (based on historical data), use the probability of default to predict the relative creditworthiness of a loan applicant. But, the credit-scoring model does not completely eliminate the human element. The selection of cutoff scores is a subjective decision. Moreover, the evaluation of applicants that have scores between the accept-scores and the reject-scores is quite subjective. Thus, to be more objective in evaluating loan applications, many institutions are exploring the use of artificial intelligence techniques such as artificial neural systems and fuzzy logic.

Fuzzy logic, a relatively new, rule-based development in artificial intelligence, tolerates imprecision and even uses it to solve problems that were not solved before. Fuzzy sets and fuzzy logic systems are based on the way the brain deals with inexact information. On the other hand, neural networks are modeled after the physical architecture of the brain. Neural networks are specialized hardware or software that emulate the processing patterns of the human brain. Fuzzy logic and neural networks are complementary technologies in the design of intelligent systems. Each method has merits and demerits. Artificial neural systems suffer from their inability to explain the steps used to make decisions and incorporate rules in their architecture. Neural fuzzy systems address some of the shortcomings of artificial neural network tools. Fuzzy logic systems often deal with issues such as reasoning on a higher level than neural networks. However, since fuzzy systems do not have much learning capability, it is difficult for a human operator to tune the fuzzy rules and membership functions from the training data set. Thus, to reap the benefits of both fuzzy systems and neural networks, a promising approach is to merge fuzzy logic and neural networks into an integrated system. Neuro-fuzzy systems represent one of the ways in which fuzzy systems and neural networks can be merged.²

This study investigates the classification accuracy of neuro-fuzzy systems to screen consumer loan applications. To construct a model, the study uses a pooled data set of consumer loan applications from nine different credit unions, and applies neuro-fuzzy system. The objective of this study is to evaluate the effectiveness of neuro-fuzzy model to identify problem loans. Secondly, we compare the performance of neuro-fuzzy model with the statistical linear discriminant analysis models. Our analysis indicates that the neuro-fuzzy models' performance in evaluating potential loan defaulters is statistically superior to the performance of linear discriminant analysis. In addition, neuro-fuzzy models do not require any restrictive assumptions of the statistical models, and are flexible enough to permit the loan officer to incorporate new rules for loan evaluation. This study is divided into six parts: Section 2 reviews the existing literature on the use of fuzzy logic and neural networks in finance, and illustrates the current applications; Section 3 describes the data used in this study; Section 4 explains the design of a neural fuzzy system model and a linear discriminant analysis model to screen consumer loan applications; Section 5 analyzes the results of loan classifications using discriminant analysis and neuro-fuzzy inference, and finally, Section 6 concludes and summarizes the study.

Section snippets

Previous studies

Many studies highlight the use of artificial neural systems and fuzzy logic in financial management. Jensen (1992) illustrates the use of a standard backpropagation neural network for loan classification. The network's classification accuracy was in the range 76–80% on the holdout sample. However, the sample size of Jensen's study is very small at 125 applicants. Tam and Kiang (1992) compare the artificial neural system approach with a linear classifier, the logistic regression model, kNN

Description of the data set

To better understand the process of credit evaluation, and explore alternative quantitative models that can aid a loan officer in the credit union environment, we analyzed data from nine credit unions. We combined the data of loans made by these credit unions to form a pooled data set with a total of 790 observations.

Discriminant analysis model

Discriminant analysis involves the linear combination of the two (or more) independent variables that differentiate best between the a priori defined groups. This is achieved by the statistical decision rule of maximizing the between-group variance relative to the within-group variance; this relationship is expressed as the ratio of between groups to within group variance. The linear combinations for a discriminant analysis are derived from an equation that takes the form of Eq. (1). $Z=W_{1} X_{1} +W_{2} X_{2}$

Multiple discriminant analysis versus artificial neuro-fuzzy inference system (ANFIS)

Table 2 summarizes the empirical results of the discriminant analysis and ANFIS.

For the seven training samples, the discriminant analysis model's predictive accuracy ranges from 69.8% to 73.2%. Groupwise, the model correctly classified in the range of 83.6–85.6% of the 250 applications in Group 1 (good credit) and 56–62% of the 250 applications in Group 2 (bad credit). Similarly, for the hold out sample, the discriminant analysis model's classification accuracy ranges between 62.05% and 66.2%

Contribution of this study

Although mathematically derived credit-scoring systems (analytic) are a definite improvement over subjective judgmental (intuitive) methods to evaluate consumer loans, we can improve the method of credit scoring considerably through the use of artificial intelligence techniques such as expert systems, neural networks, fuzzy logic, and neuro-fuzzy systems. While expert systems have been extensively used by many organizations, neural networks and fuzzy systems have captured the attention of the

Uncited references

The following works are also of interest to the reader:

Jang and Gulley, 1995, Jang and Gulley, 1998

Acknowledgements

The authors would like to thank the two anonymous reviewers for their valuable comments and suggestions. The errors in this paper are solely the authors' responsibility.

References (32)

E. Altman et al.
Corporate distress diagnosis: Comparisons using linear discriminant analysis and neural networks
Journal of Banking and Finance
(1994)
V. Desai et al.
A comparison of neural networks and linear scoring models in the credit union environment
European Journal of Operations Research
(1996)
D. Hamilton et al.
State-of-the-art-practice in knowledge-based system verification and validation
Expert Systems with Applications
(1991)
R. Lacher et al.
A neural network for classifying the financial health of a firm
European Journal of Operations Management
(1995)
M. Sugeno et al.
Structure identifications of fuzzy model
Fuzzy Sets and Systems
(1988)
G. Zhang et al.
Artificial neural networks in bankruptcy prediction: General framework and cross-validation analysis
European Journal of Operational Research
(1999)
P. Coats et al.
Recognizing financial distress patterns using a neural network tool
Financial Management
(1993)
Fabozzi, F., 1996. Bond Markets, Analysis and Strategies, Vol. 2. Prentice-Hall, Englewood Cliffs,...
Fahlman, S.E., 1988. Faster-learning variations on back propagation: An empirical study. In: Proceedings of the 1988...
R.G. Gallager
Information Theory and Reliable Communication
(1968)

D.E. Goldberg

Genetic Algorithms in Search, Optimization, and Machine Learning

(1989)

J.F. Hair et al.

Multivariate Data Analysis

(1991)

Jang, J.-S.R., Gulley, N., 1995. Fuzzy Logic ToolBox. The Math Works,...

Jang, J.-S.R., Gulley, N., 1998. Fuzzy Logic ToolBox. The Math Works,...

J.-S.R. Jang

ANFIS: Adaptive-network-based fuzzy inference systems

IEEE Transactions on Systems, Man, and Cybernetics

(1993)

J.-S.R. Jang et al.

Neuro-Fuzzy and Soft Computing, MatLab Curriculum Series

(1997)

Cited by (203)

Cost-sensitive thresholding over a two-dimensional decision region for fraud detection
2024, Information Sciences
Credit fraud poses a challenging task in terms of detection. It can result in significant losses depending on the amount, so a cost-sensitive perspective needs to be taken. Classical approaches focus on estimating the probability of fraud and selecting a decision threshold, but they often fail to consider the transaction amount or account for the cumulative losses incurred within the sample. Consequently, these approaches can result in sub-optimal strategies. A new thresholding approach is proposed, based on the construction of a two-dimensional decision space with an estimated probability and the credit amount. This expansion allows more freedom for the optimal classification rule search, which is performed with a new algorithm. The proposed method generalizes previous approaches, so an improvement is consistently achieved. In addition, it allows a restricted search. This is shown in a study of two real data sets, comparing the results obtained by a wide range of classifiers.
Credit risk evaluation using clustering based fuzzy classification method
2023, Expert Systems with Applications
Credit scoring is a crucial indicator for banks to determine the financial position and the eligibility of a client for credit. In order to assign statistical odds or probabilities to predict the risk of nonpayment in relation to many other factors, the scoring criterion becomes an important issue. The focus of this study is to propose a clustering based fuzzy classification (CBFC) method for credit risk assessment. We aim to illustrate the beneficial use of machine learning (ML) methods whose prediction power is increased by adopting fuzzy theory to calculate the default risk with a better selection of the features contributing to it. An important feature of the CBFC method is that membership values obtained as a result of the fuzzy k-means clustering algorithm are used for the purpose of better capturing the structure of an existing system.
An extensive comparison is performed to show how CBFC performs compared to the traditional ones for the datasets having different characteristics in terms of the variable types. Five different real-life datasets are studied to expose the contribution of fuzzy approach on improving the ML use. Our findings show that the proposed CBFC models can produce the promising classification results in credit risk evaluation which aid the practitioners and decision makers for issuance of credit purposes.
Bankruptcy prediction using fuzzy convolutional neural networks
2023, Research in International Business and Finance
We propose a combined method for bankruptcy prediction based on fuzzy set qualitative comparative analysis (fsQCA) and convolutional neural networks (CNN). Currently, CNNs are being applied to various fields, and in some areas are providing higher performance than traditional models. In our proposed method, a CNN uses calibrated variables from fuzzy sets to improve performance accuracy. In addition, there are no published studies on the effect of feature selection at the input level of convolutional neural networks. Therefore, this study compares four well-known feature selection methods used in financial distress prediction, (t-test, stepdisc discriminant analysis, stepwise logistic regression and partial least square discriminant analysis) to investigate their effect on classification performance. The results show that fuzzy convolutional neural networks (FCNN) lead to better performance than when using traditional methods.
Robust cost-sensitive kernel method with Blinex loss and its applications in credit risk evaluation
2021, Neural Networks
Credit risk evaluation is a crucial yet challenging problem in financial analysis. It can not only help institutions reduce risk and ensure profitability, but also improve consumers’ fair practices. The data-driven algorithms such as artificial intelligence techniques regard the evaluation as a classification problem and aim to classify transactions as default or non-default. Since non-default samples greatly outnumber default samples, it is a typical imbalanced learning problem and each class or each sample needs special treatment. Numerous data-level, algorithm-level and hybrid methods are presented, and cost-sensitive support vector machines (CSSVMs) are representative algorithm-level methods. Based on the minimization of symmetric and unbounded loss functions, CSSVMs impose higher penalties on the misclassification costs of minority instances using domain specific parameters. However, such loss functions as error measurement cannot have an obvious cost-sensitive generalization. In this paper, we propose a robust cost-sensitive kernel method with Blinex loss (CSKB), which can be applied in credit risk evaluation. By inheriting the elegant merits of Blinex loss function, i.e., asymmetry and boundedness, CSKB not only flexibly controls distinct costs for both classes, but also enjoys noise robustness. As a data-driven decision-making paradigm of credit risk evaluation, CSKB can achieve the “win-win” situation for both the financial institutions and consumers. We solve linear and nonlinear CSKB by Nesterov accelerated gradient algorithm and Pegasos algorithm respectively. Moreover, the generalization capability of CSKB is theoretically analyzed. Comprehensive experiments on synthetic, UCI and credit risk evaluation datasets demonstrate that CSKB compares more favorably than other benchmark methods in terms of various measures.
Automated detection of fibrillations and flutters based on fused feature set and ANFIS classifier
2021, Biomedical Signal Processing and Control
One of the major causes of human death worldwide is cardiac arrhythmia. Atrial Fibrillation, Atrial Flutter and Ventricular Fibrillation are considered the most prevalent among the various arrhythmia types. Hence, computer-aided detection (CAD) system has been developed to help the clinicians to interpret the electrocardiogram signal reliably within short period of time. The proposed ranking based scheme and weight factor analysis for the construction of fused feature set improve the classification accuracy compared to the latest available methods in the literature. The diagnostic ability of the CAD system has been shown in seven performance measure parameter analysis. The best classification result has been obtained using Adaptive Neuro-Fuzzy Interface System (ANFIS) classifier with the fused feature set. The proposed system of using ANFIS classifier has achieved average classification accuracy of 99.88%, precision of 99.25% and recall of 99.98%. The proposed system using ANFIS algorithm has high recognition rate compared with that of Support Vector Machine and Random Forest classifiers. Our proposed model is significantly efficient to discriminate the normal electrocardiogram signal and three cardiac arrhythmia types.
Credit Risk Evaluation of Quantum Communications Listed Companies in China Based on Fermatean Fuzzy TOPSIS
2021, Procedia Computer Science
Fuzzy set theory has been increasingly used in credit risk evaluation in recent years. This paper measures the relative closeness of 10 Chinese quantum communications listed companies to the ideal solution using Fermatean fuzzy TOPSIS method, so as to provide the credit risk evaluation of each company. It is showed that operational capabilities and technological innovation capabilities have a great impact on the company’s credit risk, as a result, for financial institutions, special attention should be paid to technological innovation capabilities when conducting risk assessments of listed companies in Quantum Communications. By comparing the algorithm used in this paper with the traditional TOPSIS method, the Fermatean fuzzy TOPSIS method achieves a higher degree of discrimination, which can be considered that a more reliable evaluation result has been obtained.

View all citing articles on Scopus

¹: Tel.: +1-610-660-3497.

View full text

Computing, Artificial Intelligence and Information TechnologyDifferentiating between good credits and bad credits using neuro-fuzzy systems

Abstract

Introduction

Section snippets

Previous studies

Description of the data set

Discriminant analysis model

Multiple discriminant analysis versus artificial neuro-fuzzy inference system (ANFIS)

Contribution of this study

Uncited references

Acknowledgements

Journal of Banking and Finance

European Journal of Operations Research

Expert Systems with Applications

European Journal of Operations Management

Fuzzy Sets and Systems

European Journal of Operational Research

Recognizing financial distress patterns using a neural network tool

Financial Management

Information Theory and Reliable Communication

Genetic Algorithms in Search, Optimization, and Machine Learning

Multivariate Data Analysis

ANFIS: Adaptive-network-based fuzzy inference systems

IEEE Transactions on Systems, Man, and Cybernetics

Neuro-Fuzzy and Soft Computing, MatLab Curriculum Series

Computing, Artificial Intelligence and Information Technology
Differentiating between good credits and bad credits using neuro-fuzzy systems