Financial default payment predictions using a hybrid of simulated annealing heuristics and extreme gradient boosting machines

Abstract

Online peer-to-peer (P2P) lending platforms face multiple challenges in today's e-commerce, but one of the most outstanding concerns evaluating loan risk based on borrowers' financial status and histories. Traditionally, financial experts assess borrowers' risk of default payments manually, but this process is tedious and time consuming, which are not widely applicable concerns for online P2P platforms. This paper proposes a hybrid of the simulated annealing and the extreme gradient boosting machine models in order to predict the likelihood of default payments based on users' finance histories. Based on the experimental results, the proposed model demonstrates its predictability with high recall scores. The proposed model not only out-performs over conventional algorithms including logistic regressions, support vector machines, random forests, and artificial neural networks, but it also provides an efficient method for optimising hyper-parameters in the machine learning algorithms. Through the utilisation of the proposed data-driven models, the necessity of tedious and potentially inaccurate human labour can be significantly reduced, and service level agreements (SLAs) can be further improved by time reduction made possible through the introduction of advanced data mining approaches.