An Application of Explainable Artificial Intelligence in Credit Scoring

Nowadays, an increasing number of decisions in business and finance are supported by artificial intelligence (AI). In modern financial technology (FinTech), especially in services such as online peer-to-peer lending or online consumer loans, approval of a loans are automatically generated by an artificial intelligence system. Most artificial intelligence algorithms based their decisions on some credit rating algorithm that evaluates the probability of default of a client. The strength of artificial intelligence is its ability to exploit multiple sources of data, many of which are non-traditional and have not been used by older statistical techniques. However, AI is highly complex, usually beyond the ability of a human being to understand. Employing black box models like AI can have severe ramifications, as the decisions made by the systems not only influence the business outcomes but can also impact many lives. Therefore, there is a dire need for explainable Artificial Intelligence or XAI which is an aggregation of methodologies and processes that facilitate humans to trust and comprehend the output and outcomes produced by machine learning algorithms. It describes an AI framework, its potential prejudices, and the expected impacts, and furthers transparency, fairness, and precision in AI-enabled decision making. In this paper, we investigate the algorithm LIME (Local Interpretable Model-Agnostic Explanations) which is a local explanation algorithm on credit-scoring datasets. Normally the decision to offer or reject a loan to a customer requires human-understandable explanation. We also provide a stability analysis of LIME on credit scoring datasets. Stability is important in decision making since that shows consistency and reasonability in the process