Overview of: “Statistical Procedures for Forecasting Criminal Behavior: A Comparative Assessment”

Abstract

Research Summary A substantial and powerful literature in statistics and computer science has clearly demonstrated that modern machine learning procedures can forecast more accurately than conventional parametric statistical models such as logistic regression. Yet, several recent studies have claimed that for criminal justice applications, forecasting accuracy is about the same. In this article, we address the apparent contradiction. Forecasting accuracy will depend on the complexity of the decision boundary. When that boundary is simple, most forecasting tools will have similar accuracy. When that boundary is complex, procedures such as machine learning, which proceed adaptively from the data, will improve forecasting accuracy, sometimes dramatically. Machine learning has other benefits as well, and effective software is readily available. Policy Implications The complexity of the decision boundary will in practice be unknown, and there can be substantial risks to gambling on simplicity. Criminal justice decision makers and other stakeholders can be seriously misled with rippling effects going well beyond the immediate offender. There seems to be no reason for continuing to rely on traditional forecasting tools such as logistic regression.

The Emergence of Machine Learning Techniques in Criminology

“…We basically agree with Berk and Bleich () on their categorization of what constitutes modern ML methods. Their selection of RF and stochastic gradient boosting for their comparative study is appropriate.…”

“…Typically, a large data set with known class membership is used to “learn” progressively a new predictive classification function that, after validation, can classify new unknown cases accurately. Predictive performance is assessed using various indicators of errors as demonstrated by Berk and Bleich (). We agree that RF and several other ML methods have great promise for supporting a wide range of criminal justice applications.…”

“…These discovery tasks use ML approaches such as K ‐means clustering, latent class mixture models, hierarchical clustering, neural networks, self‐organizing maps, density estimation methods, and so on. Some emerged decades ago (e.g., K ‐means) and clearly would not qualify for Berk and Bleich's () category of “modern ML” methods. However, most of these methods have had successive upgrades in the computer science and related ML literatures (Jain, ).…”

“…We view Berk and Bleich's (, this issue) article as potentially an important contribution to both criminology and practical criminal justice decision making. Although it is focused on new machine learning (ML) methods for forecasting, the article ramifies into several general problems that challenge the dominant, widely used standardized methods in criminology.…”

The Problem of Risk Assessment: Can Better Crime Prognoses Reduce Recidivism?

“…Computer generated models that predict where crime is going to occur are used by the police throughout the United States of America, but as yet they are not a profiling tool to identify who is committing crimes (Mangino 2013). Nevertheless, there is great confidence in the U.S. that the remaining problems can be solved by improving statistical procedures (Berk and Bleich 2013;Perry et al 2013).…”