An Adaptive Machine Learning System for predicting recurrence of child maltreatment: A routine activity theory perspective

摘要

Child maltreatment, including abuse and neglect of children, is one of the most appalling activities routinely occurring in the United States. Every year millions of child maltreatment incidents are reported to Child Protection Services (CPS) agencies. Experts in child protection face enormous workloads of analyzing reported incidents to assess the victims’ risk of experiencing a reoccurrence of maltreatment in the future. However, the existing systems deployed to help the experts are limited in two aspects: first, most have a limited capability in integrating a large amount of data originating from various entities dealing with child maltreatment; second, they are not adaptable enough to accommodate various degrees of un-structuredness inherent in the knowledge-intensive nature of CPS’s tasks, thus relying on CPS and various other experts for system configuration beyond typical users’ skillsets. In response, we propose an adaptive machine learning system (AMLS) inspired by the Routine Activity Theory focused on predicting recurrent child maltreatment. Our system offers a robust prediction for the recurrence of child maltreatment using a multi-faceted adaptive capability. We perform and report the results of extensive computational analysis to demonstrate the superiority of our system’s performance over various existing systems currently deployed.