Detecting Child Abuse Using Machine Learning Data-Driven Approaches to Enhance Early Intervention

Abstract

Child abuse is a critical issue in Bangladesh and worldwide, with devastating consequences for victims' physical and mental well-being. The need for timely detection and intervention is vital to mitigate its impact. This research aims to develop a machine learning-based predictive model to identify instances of child abuse and classify them into four categories: physical, emotional, sexual, and neglect. Using a dataset with 1,010 rows and 22 attributes, we applied various machine learning algorithms, including Logistic Regression, Random Forest Classifier, Naive Bayes, and Support Vector Machine. SMOTE (Synthetic Minority Oversampling Technique) was employed to address class imbalance, significantly improving model performance. The Random Forest Classifier demonstrated the highest accuracy (99%) with precision, recall, and F1-scores of 0.98, 0.99, and 0.98, respectively, after applying SMOTE. These results validate the effectiveness of our approach in predicting child abuse cases with high reliability. This model not only serves as a valuable tool for early intervention but also offers insights to policymakers and social workers to prioritize cases and allocate resources efficiently. Our work contributes to the growing body of research on technological solutions for child welfare, especially within the socio- cultural context of Bangladesh.