Spatial Child Maltreatment Risk Prediction Modeling: Bias and Ethical Challenges

Child maltreatment has become a prevalent problem and draws increasing concerns around the world. According to a recent study, approximately one in four children may experience child abuse or neglect in their lifetime. The impact of child maltreatment has also been linked to various long-term health concerns. The advances of machine learning and linked big data have enabled the development of new predictive spatial risk models. Predictive spatial models can assess a numeric risk score over a region of interest. The prediction results can help decision makers identify the population at risk and gaps of protective service coverage areas, thereby allowing early intervention and prevention opportunities. Family protection agencies of several states in the United States have initiated such studies.

In this talk, we present ongoing collaborations on developing a spatial risk model utilizing administrative data from heterogenous data sources. Rather than focusing on the household and child level, our approach divides a large region of interest into small spatial areal units for which a risk score can be assessed. The assessment helps decision makers in child protection services generate insights and promote prevention actions at the community level. We present the overall problem, proposed approach, and current progress of our projects. The presentation will focus on data bias and ethical challenges associated with building such predictive models.
 
Bio: Dr. Weijia Xu is a research scientist and the Manager of the Scalable Computational Intelligence group at Texas Advanced Computing Center (TACC) at the University of Texas at Austin. He received his Ph.D. in Computer Science at UT Austin and is an experienced data scientist. Weijia's main research interest is to enable data-driven discoveries through developing new computational methods and applications that facilitate the data-to-knowledge transfer process. Weijia leads the group that supports large-scale, data-driven analysis and machine learning applications using computing resources at TACC.