Credit scoring is a process to calculate the risk associated with a credit product, and it directly affects the profitability of that industry. Periodically, financial institutions apply credit scoring in various steps. The main focus of this study is to improve the predictive performance of the credit scoring model. To improve the predictive performance of the model, this study proposes a multi-layer hybrid credit scoring model. The first stage concerns pre-processing, which includes treatment for missing values, data-transformation, and reduction of irrelevant and noisy features because they may affect predictive performance of model. The second stage applies various ensemble learning approaches such as Bagging, Adaboost, etc. At the last layer, it applies ensemble classifiers approach, which combines three heterogeneous classifiers, namely: random forest (RF), logistic regression (LR), and sequential minimal optimization (SMO) approaches for classification. Further, the proposed multi-layer model is validated on various real-world credit scoring datasets.
TopIntroduction
Credit scoring is a way to determine the risk united with credit products by applying statistical or machine learning techniques on applicants’ historical data (Mester,1997). It is indicated by (Thomas, Edelman, & Crook, 2002) “Credit scoring is a set of decision models and their underlying techniques that aid credit lenders in the grantingof credit (Louzada, Ara, & Fernandes, 2016). It attempts to separate the effect of different candidates’ characteristics dependent on criminal conduct and defaults. The primary focal point of credit scoring is to pick whether a credit candidate has a place with reliable or non-financially sound group. Credit represents to the amount that is borrowed by a customer from a financial institution. Credit limit to a customer is decided by system on the basis of customer’s credentials like annual income, property and etc. Various advantages of credit scoring for credit businesses incorporate ascertaining and diminishing credit risk and cash flow improvement (Paleologo, Elisseeff, &Antonini, 2010) and its performance is accountable for the effectiveness of credit industries. It is not a single step process, periodically; financial institutions carry out it in various steps(Paleologo, Elisseeff, &Antonini, 2010),(Edla, Tripathi, Cheruku, &Kuppili, 2018) as follows:
- –
Application Scoring: It is utilized for evaluating the authenticity and suspiciousness of new candidates. That assessment is done based on social, monetary, and other information gathered while submitting the application.
- –
Behavioral Scoring: It is comparative as previous case; however it is for the current clients to investigate their personal conduct standards and to help dynamic portfolio administration processes.
- –
Collection Scoring: It categorizes customers into various groups. According to their group belongingness, banking system pays attention on those groups such as more, moderate, no etc.
- –
Fraud Detection: Fraud scoring models rank the candidates agreeing to the relative likelihood that a candidate might be unscrupulous.
Along with credit cards and home loans various credit products such as education loan, personal loan, car loan, mortgage finance, mini & micro finance etc. are also offered by financial organizations. Due to large number of new applicants and existing customers, credit scoring is not possible to do manually or it requires huge number of experts with domain knowledge and behaviors of customer. Currently, credit scoring is not limited to banking or credit industries only, various other domains such as telecommunication, real estate etc. are also applying credit score prediction models for analysis of customers’ behavior. Therefore, artificial intelligence may overcome the problem of manual credit scoring. Improving the predictive performance of model especially applicants with non-creditworthy group will have great impact for financial institution (Wang, Ma, Huang, & Xu, 2012),(Tripathi, Edla, &Cheruku,2018). This study focuses to enhance the classification performance of model by reducing the irrelevant and noisy features.
Reminder of the study is structured as follows: Section 2 describes a concise literature review, Section 3 presents proposed credit scoring model, and Section 4 exhibits the test results investigation of proposed approach along with the comparative analysis on credit scoring datasets followed by the concluding remarks based on obtained experimental results.
TopLiterature Survey
The greater part of the scientists have considered to acknowledge risk assessment as a twofold class order issue and observed it to be dependable to investigate shrouded designs in the credit scoring information. These frameworks help experts to improve their insight for credit risk assessment. In this unique circumstance, an assortment of Machine Learning (ML) strategies is used to put on view the risk assessment frameworks.Family of classifiers such as “Artificial Neural Network (ANN)” and “Support Vector Machine (SVM)” with various kernel approaches and many more classifiers have contributed significantly to improve credit risk prediction.