Exploration in Predictive Analysis and Machine Learning

From the latter half of the 20th century through the present, the scientific literature has strongly concluded that everyday decisions based on easily available data are fraught with human error, and the idea that humans are strictly rational decision- makers has come into question. Moreover, mentally processing everyday information often involves more "efficient" fast thinking at the expense of more effortful slow thinking, resulting in the use of heuristics that frequently result in error. Furthermore, the advent of large, complex systems has increased the likelihood of potential human errors in judgment due to the intricacies around extracting patterns from multifaceted data structures. One knows something about the ways in which scientists might assist decision-makers in extricating relevant patterns. One way is to inoculate decision-makers through training on the types of errors that are made in everyday decisions. Another way is by developing actuarial instruments to assist decision-makers. This chapter explores one other way of assisting decision-makers: machine learning. It begins the journey by discussing: the task of prediction; current practices and their limitations; and the promise of these newer analytic techniques. The chapter also involves describing how one arrived at the current destination. The chapter describes a study that was conducted. The goals were to provide support for caseworker decisions by predicting the likelihood of achieving permanency and the likelihood of achieving discharge within a targeted number of care days. To optimize practical utility in this study, we focused on information known at admission. Results of the study were intended to inform caseworker decision-making as early in the child or youth's time in care as possible. Participants were children and youth placed in either treatment foster care or residential care. The Child and Adolescent Needs and Strengths Instrument was used to evaluate progress towards goals to make recommendations for service planning. Results are presented by discussing the usage of 2 machine learning approaches. The relevance of this study is in offering a machine learning approach to predict foster child and youth outcomes and in the potential practical utility of the model itself. It has the potential for practical utility in that more accurate decisions hold the promise for more influence, efficiency, and accessibility. Although this work is not intended to advocate for decisions that are solely made based on machine learning models, machine learning approaches can be used to improve decision-making and lessen judgment inaccuracy due to human error. (PsycInfo Database Record (c) 2022 APA, all rights reserved)