An Optimized Light-GBM Based Classification Model for Effective Classification of Loan Defaulter

Risk analysis techniques are powerful tools that help professionals manage uncertainty and can provide valuable support for decision making. Recent techniques in the area of credit risk modeling have considered and adopted the use of artificial intelligence (AI), machine learning (ML) and deep learning (DL) algorithms. The purpose of this study is to apply an ensemble of machine learning algorithm to the classification of credit worthiness of a loan applicant. Data containing information about features associated with credit worthiness was collected from an online public data repository as a spreadsheet file that was stored in .csv format. The preprocessed dataset was split and fed to Light-GBM ensemble model to develop the classification models for credit worthiness using the holdout method over three simulation runs. The performance of each simulation run for each model was evaluated based on accuracy, recall, precision and f1-score. The study revealed that the ensemble learning model that was adopted in this study achieved very accurate results and proved to be more objective than subjective rule-based models. The results showed that there is a relative degree of importance that the features have with one another relative to the classification of credit worthiness. The study concluded that ensemble models are very effective in the classification of loan defaulter. This study recommended that future study could focus on determining the impact of feature importance on the performance of ensemble learning algorithms adopted for the classification of credit worthiness among loan applicants.