Feature selection in P2P lending for default prediction using grey wolf optimization and machine learning

Abstract

Online loan services like peer-to-peer (P2P) lending enable lenders to transact without bank intermediaries. Predicting which lenders are likely to default is crucial to avoid bankruptcy since lenders bear the risk of default. However, this task becomes challenging when the P2P lending dataset contains numer- ous features. The prediction performance could be improved if the dataset fea- tures are relevant. Hence, applying feature selection to remove redundant and irrelevant features is essential. This paper introduces a novel wrapper feature selection model to identify the optimal feature subset for predicting defaults in P2P lending. The proposed method includes two main phases: feature selection and classification. Initially, grey wolf optimization (GWO) is used to select the best features in P2P lending datasets. Then, the fitness function of GWO is as- sessed using ten different machine learning (ML) models. Experimental results indicate that the proposed model outperforms previous related work, achieving average accuracy, recall, precision, and F1-score of 96.77%, 80.73%, 97.52%, and 80.06%, respectively.