An Ensemble Method for Predicting Loan Eligibility in Commercial Bank Using Machine Learning

Abstract

In recent days in Bangladesh the number of loan applicants for loans in commercial banks are gradually increasing every year. Banking sectors always need a more accurate system for handling many issues. In order to select the right applicant who can return the loan amount within given time, the bank employees do a lot of analysis on the information provided by the applicant and based on the analysis give a prediction. But it is very difficult and time-consuming process for bank employees. To deal with this particular problem of predicting the right applicant for loan request we use the EDA (Exploratory Data Analysis) technique. A variety of machine learning models are used to aid in the task of loan prediction. A dataset made up of loan projections is used to assess the study. Data cleaning procedures were applied, such as deleting null columns, using the mean mode approach to fill in missing values, and converting categorical values to numeric format. We employ two different methods to provide the greatest outcomes from the feature selection process. Traditional machine learning models employ distinct training and testing processes for both features, which are derived from various feature selections. Bagging Classifier, out of all the models, has attained the highest level of accuracy (88.00%), as well as a high recall and F1 score. The method of univariate feature selection was used to achieve this. As a result, the results suggest that Bagging Classifier might do very well when it comes to the task of predicting loan defaults.