Machine learning techniques can help state agencies improve the distribution of their resources, directing them to the families that need them most. This article summarizes local experiences in the USA where this technology was used for a) prioritizing cases to be investigated and b) meetings with families next to leave the protection system, and c) evaluating the provision of goods and services of state agencies that are part of the system.

The Florida Department of Children and Families (FDCF) and the Allegheny County in Pennsylvania (ACP) implemented models to predict risk of abuse and fatalities in their crisis lines and thus prioritizing cases for investigation (Cuccaro-Alamin et. al 2017).

The FDCF identified 14 risk factors associated with an increased risk of death, such as the child’s age, the child having been previously removed from his or her home, or the presence of physical or mental disability. The FDCF also developed an application for social workers to monitor risk factors and intervene in time to ensure the safety of children.

The ACP assigns a risk score indicating the likelihood that the child will be separated from the family and that the case will be referred back into the system within one year of the call. They are also considering using it to prioritize families’ invitations to the support programs.  

The government of New York city used predictive models to identify risk factors among families who frequently interacted with protection services. Algorithms were applied to prioritize “exit meetings” among families who were about to exit the system and develop “risk cohorts” to adjust the performance measurement of agencies involved in providing services to these families.

The proposal does not intend to replace social workers’ professional opinion, but rather to provide them with another tool that allows them to assess risk, and to prevent children and families from not receiving the services they need due to an erroneous characterization of the risk level, which can perpetuate the situation of abuse or neglect.

In short, machine learning techniques and predictive analytics can help to better allocate the resources of the Protection System and thus protect more children and adolescents.