This study aimed to identify key risk factors and predict digital intimate partner violence (DIPV) exposure and perpetration among university students using machine learning (ML) algorithms. A cross-sectional online survey was conducted with 1,764 university students (age range = 18-41 years, M = 20.8; 87.2% female, 12.8% male) selected through snowball sampling from a large public university in Türkiye. The survey included sociodemographic, lifestyle, and relationship variables, along with the Digital Intimate Partner Violence Scale. Six ML models were used: Logistic Regression (LR), XGBoost, Gradient Boosting (GB), Random Forest (RF), LightGBM, and Support Vector Machines (SVM). Model performance was evaluated using accuracy, precision, recall, F1 score, and receiver operating characteristics-area under the curve (ROC-AUC). XGBoost achieved the highest performance (AUC = 0.996), followed closely by RF and LightGBM (AUC = 0.995). LR and GB also performed well (AUC = 0.992), while SVM had slightly lower performance (AUC = 0.989). SHapley Additive exPlanations analysis revealed that domestic violence history, urban residence, father's low education, short relationship duration, and frequent digital communication were risk factors. High income perception and non-smoking reduced DIPV risk. ML models, particularly XGBoost, effectively predict DIPV. Socioeconomic and psychosocial factors should be targeted in prevention efforts, alongside digital literacy and support services.
Özsezer et al. (Sun,) studied this question.