Unpacking Divorce: Feature-Based Machine Learning Interpretation of Sociological Patterns

Abstract

This study introduces a machine learning-based framework aimed at identifying and interpreting the most influential factors contributing to divorce. Utilizing data from the 2021 Turkey Family Structure Survey, we apply Random Forest and Logistic Regression models to rank predictors based on their relative impact on marital dissolution. The goal is to uncover which socio-legal, temporal, and behavioral variables most significantly contribute to the divorce outcome within a culturally grounded dataset. Both models converge on a set of dominant features—psychological conflict responses, cultural marital rituals, and political disagreements—demonstrating their robust influence across different algorithmic paradigms. Feature importance scores derived from model outputs and explainability tools (e.g., permutation and coefficient-based rankings) reveal consistent patterns and offer interpretable insights aligned with sociological theory. This approach contributes to computational sociology by showcasing how machine learning can be used not only for prediction, but more importantly, for identifying statistical patterns that reflect social structures and behavioral dynamics associated with divorce outcomes.