Development of a machine learning-based prediction model: Identifying high-risk households and child maltreatment risk

The present study aims to develop a machine learning (ML)-based child maltreatment prediction model using the eHaengbok-eUm dataset, a national-level social welfare data integrating administrative data dispersed across various national agencies. In doing so, it aims to predict the risk of child maltreatment through a comprehensive analysis of various factors, including household composition, household characteristics, and social risk characteristics. The findings indicate that the presence of non-kin adults in the household, parental incarceration status and age, household poverty, and number of children in the household emerged as key predictors of child maltreatment. In addition, social risk factors such as release from correctional facilities, lack of paid health insurance, and the presence of family members with disabilities, were found to increase the risk of child maltreatment. This study demonstrates the potential to prevent child maltreatment by analyzing various risk factors through a ML-based prediction model, identifying high-risk households early, and implementing customized interventions. The study provides policy recommendations for integrating the model into the social welfare system and highlights the need for practitioner training. It also underscores the importance of data ethics and personal information security to ensure sustainable use of ML-based prediction models. Overall, the findings support a multidimensional, integrated approach to enhancing child welfare and safety and demonstrate the potential of ML methods in preventing child maltreatment.