Flexible logic mining model via conditional random 2 satisfiability in discrete Hopfield neural network

The study of logic mining has gained significant attention for its ability to interpret behavior of the datasets. However, existing logic mining models face limitations such as reliance on single optimal logic, insufficient focus in attribute selection methods and a tendency to overfit especially when dealing with imbalanced datasets. To address these challenges, this paper introduces a versatile and flexible logic mining model based on Conditional Random 2 Satisfiability. The proposed model is structured into three main phases: pre-processing, learning and retrieval. By utilizing a multi-objective retrieval phase, the model improves its classification performance by retrieving a final neuron state that is optimally diversified. A different pathway in generating the best logic expands the search space leading to the production of a set of induced logics using different confusion matrixes. As a result, the logic mining model moves beyond a rigid dependency on the single best logic and thus offers flexibility in interpreting the behavior of the datasets. Additionally, the similarity analysis enhances the attribute selection method by calculating the distance between attributes and dataset outcomes. This evaluation considers the distribution of both positive and negative entries across independent and dependent attributes. Comparative experiments conducted on real-world datasets demonstrate the superiority of the proposed logic mining model, achieving an average ACC = 0.8275, PRE = 0.9116, SPE = 0.9994, MCC = 0.5221 and MCR = 0.1758. These results consistently outperform baseline logic mining models across all evaluation metrics.

Text 126793.pdf - Published Version Available under License Creative Commons Attribution. Download (13MB) Official URL or Download Paper: https://linkinghub.elsevier.com/retrieve/pii/S2773...

Actions (login required)

View Item