A Deep Learning Approach for Loan Default Prediction Using Imbalanced Dataset

Lending institutions face key challenges in making accurate predictions of loan defaults. Large sums of money given as loans are defaulted and this causes a substantial loss in business. This study addresses loan default in online peer-to-peer lending activities. Data for the study was obtained from the online lending club on the Kaggle platform. The loan status was chosen as the dependent variable and was classified discretely into “default” and “fully paid” loans. The dataset is preprocessed to eliminate all irrelevant instances. Due to the imbalanced nature of the dataset, the adaptive synthetic (ADASYN) oversampling algorithm is used to balance the data by oversampling the minority class with synthetic data instances. Deep neural network (DNN) is used for prediction. A prediction accuracy of 94.1% is realized and this emerged as the highest score from several trials with variations in batch sizes and epochs. The result of the study clearly shows that the proposed procedure is very promising.