Unsupervised Logic Mining in Classifying Divorce Outcome

Abstract

In the current era, the increasing number of divorce cases highlights the critical need to understand the behavioural causes that lead to marital breakdown. This study introduces an enhanced logic mining model developed to classify marital outcomes based on behavioural attributes from the Divorce Prediction dataset. The proposed framework integrates a satisfiability-based reverse analysis model with a neural network structure to extract interpretable logical rules from data. A similarity-based selection method is applied during preprocessing to group related attributes, improving the quality of induced logic. The model identifies optimal attribute combinations and generates clear, human-understandable rules that explain marital stability or risk of divorce. The proposed model was evaluated on the real-world Divorce Dataset composed of psychological and behavioural indicators. The results show that the proposed framework effectively improves logic interpretability and classification performance compared to existing methods based on performance metrices. This study contributes to the development of explainable logic-based models that can support counsellors, researchers, and policymakers in understanding behavioural patterns associated with divorce outcomes. However, the model has been tested only on a single behavioural dataset, which may limit its generalizability. Future research could extend this approach to other social and psychological domains to further validate its performance.