By Ellie Sams
Kevin Jones is no stranger to innovation. Under his leadership as CIO, the Indiana Department of Child Services (DCS) has embarked on two huge IT modernization efforts for both Child Support & Child Welfare. Indiana DCS continues their innovation, with the support of Thoughtspot and Snowflake, by hosting a hackathon to encourage the brightest tech minds to develop solutions to help solve their unique challenges.
Indiana Department of Child Services (DCS) faces a large number of false positive child abuse allegations.
Allegations of child abuse are reported through a hotline 24 hours a day, 7 days a week. In 2019, there were 121,726 reports of allegations. Each report requires a DCS family case manager to become involved. The average response time for reports is 2 hours to 5 days and the average length of assessment is 40 days.
The hotline recommends a course of action after completing the Structured Decision Making in the intake tool. An internal DCS Research & Evaluation team found the agency’s intake system is risk averse, resulting in high false positive rates. A false positive means that a report was screened in, unsubstantiated, and with no subsequent reports within the following 180 days. This highlights the concerns that agency resources may be inappropriately directed towards investigation and away from intervention, in turn necessitating testable decision-making changes to reduce false positive errors.
Reducing false positives allows for case managers to focus their time on correct investigations, reduces workload on already stretched resources, and avoids unnecessary stress on children and families.
The full problem statement can be read here.
The team considered many of the unanswered questions in order to solve the problem. Our goal was to leverage our Analytics & AI expertise to help DCS identify false positive cases as early as possible in the process. Cardinality wanted to ensure we defined key actions that could substantially accelerate the determination of investigating the right cases.
The team took a three-pronged approach to develop and execute the solution:
We leveraged the Random Forest Classifier algorithm to build the model based on assessment data. The model identified ~10 out of the 37 questions to be statistically significant in classifying a case as False Positive or not with an accuracy of ~90%. We have proposed a Smarter Decision Engine which leverages this model to predict the Screen In or Screen Out recommendations for any reported case.
DCS also can leverage AI & ML models to remove duplicate people from internal systems to have high quality data to perform analysis. Advanced modelling techniques can be used to compute a statistical likelihood that multiple records are relating to the same person.
The recommended solution allows Indiana DCS to leverage data across other State & Federal Child Welfare systems such as NCANDS, LONGSCAN, SACWIS, AFCARS, and CFSR to develop a holistic understanding around child abuse & care.
The research, development, and design Cardinality presented in the hackathon gives Indiana DCS the opportunity to provide better training to hotline workers, ensuring comprehensive and correct data collection. The solution also enables Indiana DCS to improve their decision making process because of intelligence provided by the Classifier Model. Cardinality-ai is making it easier for Indiana DCS to have a universal understanding of the factors that have contributed to false allegations of child abuse.
Not only will the solution presented by Cardinality improve and implement improved processes, but it will also solve the biggest problem: easing the burden off caseworkers and providing support to children and families in need.