Multi-unit discrete hopfield neural network for higher order supervised learning through logic mining: optimal performance design and attribute selection

In the perspective of logic mining, the attribute selection, and the objective function of the best logic is the two main factors that identifies the effectiveness of our proposed logic mining model. The non-significant attributes selected will cause the Discrete Hopfield Neural Network to learned and obtain wrong synaptic weight. Thus, this will result to suboptimal solution. Although we might select the correct attributes, the conventional objective function of the best logic limits the search space to obtained more induced logic during the retrieval phase of Discrete Hopfield Neural Network. Therefore, this paper proposes a novel logic mining by integrating statistical analysis in the pre-processing phase to ensure that only optimal attributes will be selected. Supervised learning approach via correlation analysis is implemented for the purpose of attribute selection. Additionally, permutation operator serves to enhance the probability of the higher order satisfiability logical rule to be satisfied by having finite arrangement of attributes. During the learning phase, we proposed multi-unit Discrete Hopfield Neural Network to enhance the search space which leads to optimal solution. The efficiency of the proposed model is tested on 15 real-life datasets by comparing the performance of the model with existing works in logic mining using five performance metrics including accuracy, sensitivity, precision, Matthews Correlation Coefficient (MCC) and F1 Score. According to the results, the proposed model has its own strength by dominating most of the average rank of the performance metrics. This demonstrates that the proposed model can differentiate across all domains in the confusion matrix. Additionally, the p-value obtained based on the five-performance metrics indicate that there is a significantly difference between the proposed model and all existing works since the value obtained for accuracy (0.000), sensitivity (0.001), precision (0.000), F1 score (0.000) and MCC (0.000) are less than 0.05. This finding statistically prove that the proposed model is more effective compared with existing works in logic mining.