<![CDATA[Divorce is a social problem that continues to increase and has negative impacts on the psychological, social, and economic conditions of individuals and families. This study aims to build a divorce risk prediction model using the Random Forest algorithm with hyperparameter optimization using the Random Search method. The dataset was obtained from the Kaggle platform with 170 samples and 54 psychological-behavioral attributes of couples. The research stages included data preprocessing, dataset splitting (80:20), baseline model development, hyperparameter optimization with Random Search, and evaluation using accuracy, precision, recall, and AUC-ROC metrics. The results showed that the model achieved 94.12% accuracy on the test data with 97% recall that minimizes false negative risk. Hyperparameter optimization successfully improved the model's internal stability with a cross-validation average of 98.57%, although the test accuracy was equivalent to the baseline model. A gap of 4.45% between validation and test accuracy indicates potential overfitting, which is common in small datasets. Feature importance analysis revealed five dominant psychological factors: willingness to compromise, effective communication, conflict resolution, alignment of life values, and forgiveness ability. This research contributes to the development of an early detection system for divorce risk based on machine learning and provides an empirical basis for more targeted counseling interventions]]>
Copyrights © 2026
