Blockchain Integrated Data Slicing based Model Validation for AI-ML Credit Risk Management Systems

Credit risk assessment or management is a process of evaluating risks on customers' debt obligations and assessing whether the customer may default on debts, which are considered the finance industry's cornerstone.The main objective of this assessment is to examine good and bad debts.Assessing and managing such credit risks in banking and other businesses offering credit to customers is the most crucial problem.It has been addressed in several ways in the literature, such as i) Qualitative approaches -assessment is based managers' experience, ii) Statistical methods in which historical data is analysed for its relationship between customer characteristics and their default status, iii) Machine learning and Artificial Intelligence models which handle complex data and nonlinear relationship that exist among the data and iv) Ensembling models in which strengths of various models are combined.Among these, traditional ML models, Advanced AI-ML models, and Ensembling Models are the most adopted due to their higher levels of predictive accuracy.The AI model for credit risk assessment implements machine learning algorithms to analyze and predict credit risks in an automated way.Machine learning algorithms offer significant benefits, such as accuracy and the ability to handle complex data.However, these ML-based implementations introduce critical challenges, such as validation of model fairness, algorithmic bias, lack of transparency, etc.Also, AI-based models are black boxes, so it is very hard to analyse why the system made such a decision.Interpreting such ML algorithms will help us understand the model's weaker points towards the different data segments and enable us to handle those data segments separately to improve the efficiency of the credit risk assessment process.Hence, there is a need for a data slice-based evaluation model for performance.This paper proposes a novel framework of a slicebased validation model for an AI-ML enabled credit risk assessment model, which allows granular assessment of the AI-ML model and integrates blockchain, which provides an immutable, transparent ledger for storing slice-based validation metrics, model version, etc.The results show the increased efficiency of the proposed credit risk assessment model.