Loan Default Forecasting Using StackNet

Credit risk analysis is a process used by financial institutions to estimate the creditworthiness of the borrower. Financial institutions do this to protect their revenues against loan default. Peer to peer lending is a growing and popular microfinance tool used by financial institutions to provide their customers with an online platform to match borrowers and lenders. This way of matching lenders with borrowers happens with the use of some algorithm. Unfortunately, no algorithm is foolproof, and where there is a process of loan or lending money there is a risk of loan default. Financial institutions try to solve this problem using various credit risk models. This paper provides a credit risk model which predicts loan default using a stacked ensemble model. The dataset used for this research is of a leading European P2P application called Bondora which is available on Kaggle. Our stacked model will consist of AdaBoost, XGBoost and Random Forest using StackNet framework. StackNet implementation helped in increasing the performance of our credit risk model. The StackNet classifier achieves an AOC-ROC score of 99.4% and accuracy of 96.9% when using 15 features. This proves StackNet classifier demonstrates excellent performance in forecasting default in P2P lending.