Predicting Elder Abuse Using a Random Forest Classifier

This paper aims to identify key factors influencing elder abuse within the family and to further explore the heterogeneity of these factors across different types of elder abuse. The data were drawn from the China Longitudinal Aging Social Survey and the final valid sample was 10,703. A random forest classifier, a supervised machine learning method, was employed to identify the key influencing factors of elder abuse. Children's economic status is found to be the most important factor in predicting elder abuse, followed by the number of children, the health status of older people, intergenerational relations, children's time pressure, and the provision of home-based elderly care services. The likelihood of elder abuse decreases continuously with better economic status and less time pressure of children, better health of older people, more children, and better intergenerational relationships, whereas the influences of home-based elderly care services on elder abuse are not monotonous. Number of children contributes most to predict financial abuse, while children's economic status plays the most significant role in predicting physical and psychological abuse and neglect. This study is the first to apply a supervised machine learning approach with a random forest classifier for the identification of risk factors associated with elder abuse. The findings highlight the advantages of machine learning techniques in improving the prediction accuracy of elder abuse compared to traditional econometric models.