Abstract

Child sexual abuse is a pervasive and distressing issue that poses serious threats to the well-being and development of children. Early identification and prevention of such incidents are crucial for ensuring child safety and protection. In this study, we investigate the application of stacked machine learning models for the forecasting of child sexual abuse cases. Data on child sexual abuse incidents were gathered from StatBank Denmark and used in this analysis. The geographical coordinates of the municipalities were incorporated as part of the descriptive analysis to examine the distribution and prevalence of child abuse cases. Our approach incorporates a stacked ensemble framework that combines the XGBoost, LSTM, and Random Forest algorithms. By leveraging the strength of individual models and capturing diverse patterns in the data, the stacked model aims to improve prediction performance. Our experimental results demonstrate that the CSA-Forecaster model outperforms individual models in forecasting child sexual abuse incidents. The proposed model achieved an RMSE of 0.094, MAE of 0.0712, MAPE of 0.1557, and R2 of 0.8028, indicating robust performance. The outcomes of this research have significant repercussions for the creation of proactive interventions and support systems. Child protection agencies and experts might be equipped to more effectively allocate resources and potentially prevent future abuse instances by employing machine learning models.