PREDICTING ACCEPTANCE OF THE BANK LOAN OFFERS BY USING SUPPORT VECTOR MACHINES

Abstract

Loans are a crucial profit source for banks, which try to identify trustworthy customers for personal loan
offers. However, these offers can sometimes be declined by customers. Predicting which customers will accept loan
offers adds extra work for banks, but accurate predictions can enhance profitability. This study aims to forecast the
acceptance of bank loan offers using the Support Vector Machine (SVM) algorithm. SVM was used with four
different kernels, utilizing a grid search algorithm for optimal predictions and cross-validation for increased
reliability. The research findings reveal that the polynomial kernel achieved the highest accuracy at 97.2%, while
the sigmoid kernel had the lowest accuracy at 83.3%. Due to the unbalanced dataset, with a ratio of 1 positive to 9
negative instances, some precision and recall values were notably low, such as 0.108 and 0.008, respectively. This
study recommends the use of SVM in banking systems for predicting the acceptance of bank loan offers. Loans are
one of the major sources of income in the gadget industry. Banks try to choose reliable clients and provide them
with non-public loans, but clients can sometimes be refused bank loans. Predicting this problem is more work for
banks, but if they can predict that customers will get a personal loan, they can make more money. Therefore, at
present, the purpose of this review is to confirm the bank's credit rating using the Support Vector
Machine (SVM) algorithm. In this context, SVM is used to consider the effects with the four bases of SVM, as well as
grid search rules for better prediction and again convey the guarantee of all good results.