The conventional Discrete Hopfield Neural Network encounters a notable challenge in generating an output representation that is interpretable by the user. In response to this challenge, a symbolic rule has been introduced to govern the information embedded in the network. This approach has proven successful, leading us to develop a logic mining model that utilizes the logical rule of 3 Satisfiability in Discrete Hopfield Neural Network to represent attributes for repository datasets. Nevertheless, the existing 3 Satisfiability Reverse Analysis model faces two primary issues: random attribute selection and predetermined attribute arrangement. These issues can significantly impact the ability of the model to retrieve the optimal induced logic. In response, a solution that involves a supervised attribute selection benchmark using correlation analysis is proposed. Additionally, a permutation operator to allow for various attribute arrangements was implemented, thereby expanding the search space and increasing the likelihood of finding an optimal solution. Furthermore, a novel objective function for determining the best logic, which considers both true positives and true negatives is also introduced. This differs from the conventional 3 Satisfiability Reverse Analysis method, which relies solely on true positives. Three performance metrics, including accuracy, precision, and Matthews Correlation Coefficient (MCC), and tested on 13 real-life datasets to validate the efficiency of our proposed model. The results clearly demonstrated that our proposed model consistently outperforms the conventional 3 Satisfiability Reverse Analysis method, achieving the highest values for all performance metrics.
