Domestic violence is a social and public health problem with serious consequences for victims, as it causes physical, emotional, and psychological harm, and often leads to long-term effects that impact family dynamics and mental health. The objective of this study was to implement a predictive model for classifying the severity of domestic violence cases in Peru using Machine Learning techniques. The methodology was based on five phases: data acquisition from the NotiWeb system of the Ministry of Health (CDC-MINSA); preprocessing (data cleaning, transformation, and standardization); class balancing using SMOTE; implementation of eight supervised Machine Learning models (Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, K-Nearest Neighbors, Support Vector Machine, Bagging, and AdaBoost); and model evaluation using five performance metrics. Multiple models achieved optimal performance, with Logistic Regression (LR), Random Forest (RF), and AdaBoost (ABC) consistently showing the best results across both training/test splits. LR demonstrated the most stable performance with Accuracy, Precision, Recall, F1-Score, and AUC-ROC of 0.9540, 0.9669, 0.9396, 0.9530, and 0.97, respectively. These results demonstrate the potential of predictive algorithms for early identification and prioritization of high-severity domestic violence cases, contributing to more effective institutional responses and public policy design.