Explainable Multistage Ensemble 1D Convolutional Neural Network for Trust Worthy Credit Decision

Abstract

Banking is a dynamic industry that places significant importance on risk management, requiring accurate and interpretable AI models to make transparent lending decisions. This study introduces a groundbreaking approach that combines a multistage ensemble technique with a 1D convolutional neural network (CNN) architecture. The algorithm not only delivers superior classification performance but also offers interpretable explanations for its decisions. The algorithm is designed with multiple strategic steps to enhance model performance without sacrificing explainability. Thorough experiments were conducted using datasets from private banks and non-banking financial companies (NBFCs) in India to evaluate the algorithm's performance. It was compared against various state-of-the-art models, demonstrating remarkable precision, recall, F1 score, and accuracy values of 0.994, 0.992, 0.993, and 0.991, respectively. This outperformed competing models like homogeneous deep ensembles, 1D CNN, and Artificial Neural Networks (ANN). Furthermore, individual borrower dataset evaluations confirmed the proposed algorithm's consistency and efficiency, achieving precision, recall, F1 score, and accuracy values of 0.960, 0.961, 0.952, and 0.964, respectively. The research emphasizes the effectiveness of the explanatory integration decision process, wherein the Explainable Multistage Ensemble 1D CNN not only provides enhanced credit risk prediction but also facilitates transparent and interpretable lending decisions. The algorithm's ability to offer understandable explanations empowers financial institutions to make well-informed lending decisions, reduce credit risk, and foster a more stable and inclusive financial ecosystem.