KMAN Publication Inc. (KMANPUB) Applied Family Therapy Journal (AFTJ) 3041-8798 6 Serial Number 30 2025 11 01 Early Detection of Family Violence Risk Using Ensemble Machine Learning on Psychosocial and Demographic Indicators Early Detection of Family Violence Risk Using Ensemble Machine Learning on Psychosocial and Demographic Indicators 1 9 10.61838/kman.aftj.4952 EN Journal Article 2024 05 27 <p><strong>Objective:</strong> The objective of this study was to develop and validate an ensemble machine learning framework for the early detection of family violence risk based on psychosocial and demographic indicators among South African households.</p> <p><strong>Methods and Materials: </strong>This cross-sectional predictive study recruited 1,198 adults from urban, peri-urban, and rural regions of South Africa using stratified multistage sampling. Participants completed a comprehensive assessment battery measuring demographic characteristics, socioeconomic conditions, psychosocial functioning, relational dynamics, and behavioral risk indicators. Data preprocessing included cleaning, normalization, multiple imputation, and feature engineering. The dataset was partitioned into training and test sets, and class imbalance was addressed using synthetic oversampling. Four supervised machine learning models—random forest, gradient boosting, extreme gradient boosting, and support vector machine—were trained using five-fold cross-validation and Bayesian hyperparameter optimization. An ensemble model integrating these classifiers was constructed and evaluated using accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve.</p> <p><strong>Findings: </strong>The ensemble model significantly outperformed all individual classifiers, achieving an accuracy of 0.917, precision of 0.904, recall of 0.896, F1-score of 0.900, and AUC of 0.958. Inferential comparisons indicated statistically meaningful improvements in sensitivity and discrimination over the strongest single model. Feature importance analysis revealed that family conflict frequency, emotional regulation difficulty, perceived stress, substance use severity, depression symptoms, and parenting stress were the most influential predictors of violence risk.</p> <p><strong>Conclusion: </strong>The findings demonstrate that ensemble machine learning provides a highly effective and interpretable framework for early identification of family violence risk, offering substantial potential for proactive prevention and targeted intervention in high-vulnerability populations.</p> https://journals.kmanpub.com/index.php/aftj/article/download/4952/8914