Credit Scoring in Peer-to-Peer Lending Using Machine Learning Techniques

This thesis explores the shift in credit risk evaluation of peer-to-peer (P2P) lending platforms and machine learning methods. Traditional banks face limitations in flexibility and prediction accuracy. The study reviews the evolution of P2P lending, emphasizing that incorporating socio-economic factors and addressing regulatory challenges can enhance credit risk assessment. A key suggestion is to harmonize P2P processes and shift focus from credit scoring to profit scoring to align with investor objectives. The thesis highlights the potential of non-traditional data, such as digital footprints, in expanding credit access and discusses concerns related to algorithmic bias and ethics in automated lending. Additionally, the research compares conventional and machine learning methods for credit risk assessment, concluding that data-driven algorithms can improve financial inclusion and efficiency. The practical analysis utilizes a dataset from the Bondora platform, where various machine learning algorithms, including logistic regression, decision trees, random forests, gradient boosting, and artificial neural networks. In this case, the gradient boosting model performed the best.