← All publications

Evaluating AI-Driven Risk Scoring Models for Child Protective Services Case Prioritization: Opportunities and Ethical Guardrails

Senthil Babu Malli Jayaraj
IT Project Manager, USA

International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering (IJAREEIE), February 2026.

Abstract

Child protective services (CPS) agencies increasingly consider AI-driven risk scoring models to support intake screening and case prioritization decisions, informed by prominent implementations such as the Allegheny Family Screening Tool and subsequent tools adopted by other jurisdictions. These models offer the prospect of more consistent, evidence-informed prioritization at a time when intake call volume routinely exceeds caseworker capacity to fully investigate every referral with equal depth. At the same time, published research has raised substantial concerns regarding disparate impact, transparency, and due process when algorithmic scores influence decisions with such direct consequence for children and families. This research article presents a governance framework for evaluating AI-driven risk scoring models in CPS case prioritization, examining model design and validation, disparate impact assessment, transparency and explainability requirements, human oversight design, and a five stage governance maturity model. The article draws on published algorithmic accountability research, child welfare practice literature, and prior responsible AI governance frameworks in this research series. Findings are illustrated with a distinct set of three-dimensional visualizations, including a 3D spiral scatter plot of risk score escalation across repeated referrals, a 3D binned density surface relating score to outcome, a 3D network graph of risk factor relationships, and a 3D stacked bar chart of subgroup composition by risk tier, alongside an extensive set of grid-style data tables. Drawing on published child welfare algorithmic accountability research, NIST AI risk management guidance, and responsible AI governance literature through January 2026, this article finds that agencies applying structured disparate impact assessment, meaningful human oversight, and transparent case-level explanation report more defensible prioritization outcomes than agencies deploying scoring models without these governance practices, though published research also documents cases in which governance gaps have led to serious equity and due process concerns.

Download full text (PDF)

Originally published in IJAREEIE. View the journal version.