Exploration in Predictive Analysis and Machine Learning

Abstract

This chapter explores the potential for using machine learning methods to support foster care workers by improving the accuracy of the complex but critical decisions they make on behalf of children, youth, and families. By using historical data to predict the likelihood of future events for foster children and youth as early in their time in care as possible, models can alert workers to bring services to bear that can help to avoid negative outcomes. This chapter uses the author team’s journey to compare traditional and machine learning methods. They describe the process that began with attempts to apply traditional statistical methods to the task of prediction and resulted in two machine learning models that are used as examples. Using these models, clinicians can consider applying machine learning techniques to complex historical datasets to predict foster care outcomes and support decision-making.