Automated Decision Making System for Bank Loan Approval with Real-Life Dataset

Banks are making a vital role in profits through loans. Every day, banks receive a large number of loan application forms from customers and other individuals but not all of them are approved. Manually selecting a loan application after verifying and evaluating the applicant’s eligibility is cumbersome and time-exhausting. To deal with these issues, an Extreme Gradient Boosting(XGBoost) based automated loan approval system is proposed along with real-life dataset generation. An artificial dataset mirroring the financial institution’s CBS database with 31 columns and 107,540 rows is created. To increase the system’s complexity, we intentionally added noisy data to the dataset. The Synthetic Minority Oversampling Technique (SMOTE) is used to balance the classes (approved and rejected) with a ratio of 0.882. In this study, XGBoost is used to extract patterns from an improvised real-life synthetic dataset and to predict deserving loan applicants by considering factors such as applicant salary, income eligibility, Credit Information Bureau India Limited (CIBIL) eligibility, and Non-Performing Asset(NPA) classification. The proposed scheme is helpful for bank officials to reduce the manual workload and also reduces the time for loan sanction. Also, various Machine learning models including Support Vector Machine(SVM), Random Forest, CatBoost, and LightGBM are analyzed on the proposed synthetic customer dataset. Performance evaluation through various quantifying parameters has indicated that XGBoost based technique works better than other algorithms, with a best accuracy of 96.61%.