A Systematic Survey of Automatic Loan Approval System Based on Machine Learning

Introduction

The Indian banking sector is a strong, well- capitalized, and well-regulated industry. Despite the fact that the government has been injecting capital into public sector banks via recapitalization bonds for the past two years, liquidity has become a major concern following the Covid outbreak. The banking sector is quietly struggling to meet the problems it faces, which include preserving capital sufficiency, asset quality, and growth. Bad loans are one of the major issues. Loan recovery is one of the issue that has harmed the banking sector, which is struggling to preserve the quality of its assets. In order to avoid wrongdoings, proper examination and severe application methods are required in loan approval process.

Banks receive numerous amount of loan applications on daily basis. Banks would lose money if loan repayments are delayed, thus they must carefully examine the loan approval procedure. When they first authorize a loan, they do a lot of paperwork, which results in a lot of data. Banking has improved its study of determining the potential of risk through client profile, prior expenditures, and consumer transaction history, among other things, in recent years. Yet we can see assessing the loan related risk is still one of the primary concern for every banks. It usually involves steps portrayed in Figure 1.

Figure 1.

Existing method in banks

During the process of loan approval there are difficulties on both ends of the transaction whether it is the lender or the borrower. The bank staff checks the customer profile thoroughly on a variety of parameters. They mainly checks for the default risks that should be as low as possible. The applicant must have made timely payments on past loans and should possess good credit score. As a result, loan processing takes time. One requires a good Cibil score to get their loan approved. Cibil score depicts the credit profile of an applicant. It is a three digit numeric summary of credit history having predetermined range usually in between 300 and 900. The necessity of maintaining credit history is visible in the whole process. There comes a situation when applicant is fresh without having any credit history, in such instances, there is high possibility of rejection of loan.

Technology inclusion in banking sector is one of the thing which can work out in longer term. Thus motivated by the recent advancement in automation process and problems being faced by banking industry a model is proposed for automating the loan approval system. Many researches provided solution to this particular problem using different algorithms. The aim of this study is to perform detailed analyses of previous research works to find their limitations, to extract patterns among the types of classifiers applied, dataset used and maximum efficiencies achieved by each algorithms. Findings of literature survey are used to select algorithms and dataset for proposed model. In this paper four classification algorithms Logistic Regression, Random Forest, Support Vector Machine and Gradient Boosting are applied simultaneously.

The summarizing details of the findings of this work are as follows:

• •

  Theoretical Background provide details of Indian banking system and classification of bank loans. Applications of Machine Learning, Data Analytics and Predictive Analysis are also discussed in reference to banking sector issues.

• •

  Research Approach section presents analysis and results of methodologies used in providing solution to the loan prediction problem. Results of findings are illustrated using different charts.

• •

  In Related Work section, summary of the datasets and algorithms used, with accuracies and limitations is provided.

• •

  Proposed Methodology section discusses about various classification algorithms and evaluation techniques which can be applied.

• •

  Results and Discussion section provides outcomes of exploratory data analysis and accuracies achieved by the proposed model using classification reports.

• •

  Conclusion and Future work section discusses about the results and limitations of this work and provides suggestions that can be applied in future studies.