Explaining the Artificial Neural Network Using Evolutionary Fuzzy Association Rule Mining (EFARM)

The Artificial Neural Network (ANN) is a widely used machine learning technique for various classification problems. However, despite its popularity, ANN has certain limitations requiring careful consideration. One primary drawback of ANN is its classification as a black-box model, meaning it generates outputs without explaining them. This paper introduces an innovative approach that presents an evolutionary fuzzy association rules-based explanation model to address this issue. In this proposed model, the fuzzy association rules mining technique generates refined rules for the ANN. These rules are then further optimized using an evolutionary algorithm known as NSGA-2, explicitly focusing on enhancing the model's fidelity, coverage, and rule count. To evaluate the effectiveness of the proposed model, comprehensive testing has been conducted on three benchmark datasets: Pima, WBCD, and Austra. Furthermore, a comparative analysis with four other works within the same field has been performed to assess its performance. The results demonstrate that the proposed explanation model achieves an impressive accuracy rate ranging from 92 to 100%. By offering interpretable and refined rules, the proposed approach enhances the transparency and comprehensibility of the ANN's decision-making process.