USC CAIS Associate Director Jordan Davis introduces Dr. John Prindle.
Research Assistant Professor John Prindle presented his team’s work using machine learning to predict re-occurrence of maltreated children being referred again to child welfare systems. There is a strong need to create an automated, predictive tool as a decision aid for social service providers to screen and triage children who are victims of abuse and/or neglect.
The machine learning predictive model incorporates historical data using several predictors including demographics, prior foster care placements and cases, current allegations, parental characteristics, and case histories from other agencies. The dataset that was utilized included 3.4 million observations for 1.2 million unique children, and over 300 predictors between 2010-2014. Using random forests, the model was found to be more accurate than the standard Structured Decision Making (SDM) assessment. This is because more variables were included in the predictive model than the standard tool.
For more information about the proof of concept for the predictive model, please visit: datanetwork.org
