Exploring the Potential of Learning in Credit Scoring Models for Alternative Lending Platforms

Abstract

Credit scoring is a crucial financial task that has been studied using both statistical and Artificial Intelligence (AI) methods. The biggest risk to any bank or financial institution in terms of impact is credit risk. Since credit risk strategy sets pricing and may even have an impact on apparently unrelated areas like marketing and decision-making, accurate credit risk assessment has an impact on an organization's balance sheet and income statement. Traditional financial organisations have been using credit scoring algorithms extensively for a long time. There are restrictions when using these models in P2P lending. Initially, P2P credit data typically consists of sparse category categories and dense numerical information. Second, it is typically not possible to update the credit scoring algorithms that are now in use online. P2P lending involves a high volume of loan transactions, and the distribution of data changes in response to new information. When a credit scoring model is used without taking data updates into account, significant deviations or even failures in later credit assessments result. We provide a novel Online Integrated Credit Scoring Model (OICSM) for P2P Lending in this research. Gradient boosting decision tree models and neural networks are integrated by OICSM to improve the credit scoring model's ability to handle two different kinds of features and update live. Experiments are carried out both offline and online, utilising authentic and representative credit datasets to confirm the efficacy and superiority of the suggested model. According to experimental data, OICSM's advantage in deep learning over two features may greatly increase performance, and its live dynamic update capacity can further rectify model degradation.