Prediction of Violence Against Women Using Ensemble Learning Models: A Comparative Study of LightGBM, XGBoost, and Others

Warnars, Harco Leslie Hendric Spits; Sunge, Aswan Supriyadi; Suzanna; Bevlyadi, Beni; Muyeba, Maybin K.

Abstract

Violence against women and the possibility of its occurrence among children is a very serious issue that negatively impacts the physical, psychological, and emotional aspects of the victims and those around them. Various efforts have been made to reduce violence against women and children; however, in reality, such violence still occurs significantly in many countries due to emotions and turmoil within human relationships. It is necessary to propose prediction methods so that violence can be reduced through early observation and intervention against violence experienced by women. machine learning, as one of the Artificial Intelligence algorithms, offers a solution to identify and predict the risk of violence. This study aims to explore the use of several Ensemble Learning models, such as LightGBM, XGBoost, CatBoost, and AutoEnsemble, which are expected to improve prediction accuracy and stability. This study uses a dataset consisting of 348 samples with 5 selected features that represent indicators relevant to the risk of violence against women. The test results show that XGBoost and CatBoost achieved the highest accuracy, approximately 73%, with a precision of 76%, recall of 65%, and F1-Score of 70%. TabNet demonstrated similar performance with an accuracy of 73%, but with a higher recall of 70%. Meanwhile, LightGBM showed slightly lower performance with 68% accuracy and an F1-Score of 64%. AutoEnsemble produced stable results with 73% accuracy, 76% precision, 65% recall, and 70% F1-Score. However, the practical limitation of this study lies in the relatively small dataset size, which may affect the model’s generalization ability when applied to larger or more diverse features. The findings of this study indicate that Ensemble Learning models can provide accurate and effective results in predicting violence against women. It is hoped that this research can contribute to more proactive and accurate efforts to prevent violence against women in the future.

• Journal Article Type: Article
• Acceptance Date: Mar 3, 2025
• Online Publication Date: Apr 30, 2025
• Publication Date: Apr 30, 2025
• Deposit Date: Jul 4, 2025
• Publicly Available Date: Jul 7, 2025
• Journal: International Journal of Safety and Security Engineering
• Print ISSN: 2041-9031
• Electronic ISSN: 2041-904X
• Publisher: WIT Press
• Peer Reviewed: Peer Reviewed
• Volume: 15
• Issue: 4
• Pages: 727-735
• DOI: https://doi.org/10.18280/ijsse.150409