A hybrid error correction approach for prediction of credit approval: An explainable artificial intelligence approach

One of the main issues faced in classification problems is the lack of prediction accuracy. This can be due to the use of unsuitable features for certain types of data. It is important to establish a framework that can identify the most important features, even if that means reducing the number of features. However, reducing the number of features can also lead to more prediction errors, so it is crucial to develop an approach that can select features and correct errors simultaneously. This is especially important in processing financial data and making financial decisions, as any errors can have major consequences. The paper introduces a heuristic approach that utilizes eXplainable Artificial Intelligence (XAI) in order to address this issue. Specifically, the paper uses the SHapley Additive exPlanations (SHAP) method, which is designed to identify and predict problems with credit scoring and approval. The main objective of this approach is to correct errors that may occur in conventional classification methods. The results show that, despite strict parameter tuning and selecting primary models with the highest accuracy, the proposed method has effectively corrected classification errors. The findings also demonstrate that this approach can improve error correction rates and model accuracy when applied to datasets such as the German Statlog and Australian Credit Approval datasets.