From Prediction to Policy Simulation: AI-Enabled Decision Support for Child Welfare Service Allocation

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

In social welfare systems, agencies are often required to make policy decisions under conditions of uncertainty, limited resources, and significant human consequences. The current study aims to examine how AI-enabled policy simulation can support these policy decisions in the context of identifying and preventing the recurrence of child abuse and neglect (RCAN). While prior predictive analytics research, including Han et al. (2021), demonstrates the value of data-driven approaches for identifying high-risk cases, less attention has been given to how these approaches can support broader policy questions related to service planning and resource allocation, and the potential effects of interventions across communities over time. Building on Information Systems research that frames AI as an organizational and sociotechnical challenge (Berente et al., 2021), the current project aims to develop a simulation model using Physics-Informed Neural Networks (PINNs) (Qian, 2025). The simulation will forecast RCAN risk across space and time under alternative service-intervention scenarios. Unlike traditional risk models, PINNs combine deep learning with social dynamics encoded as partial differential equations (PDEs), allowing the model to simulate intervention portfolios impact on RCAN risk across communities over time. The study will use data from the National Data Archive on Child Abuse and Neglect (NCANDS), which includes child welfare case data collected from 2001 to 2023 (Han et al., 2021; Kim et al., 2017). The analysis will focus on post-investigation services, including family preservation, legal and court-appointed services, counseling and mental health support, daycare, housing, home health services, and removal or foster care interventions. This research contributes to Information Systems literature by extending AI-enabled decision support from prediction to policy simulation and resource allocation in a high-stakes public service setting. Practically, it aims to provide child welfare leaders with a “policy sandbox” for comparing intervention portfolios before implementation, , supporting more informed social welfare decision-making.

Links and sources

• PDF from original source

Comments