Research has shown that police officer involved shootings, misconduct and excessive use of force complaints exhibit network effects, where officers are at greater risk of being involved in these incidents when they socialize with officers who have a history of use of force and misconduct. In this work, we first construct a network survival model for the time-to-event of use of force incidents involving new police trainees. The model includes network effects of the diffusion of risk from field training officer (FTO) to trainee. We then introduce a network rewiring algorithm to maximize the expected time to use of force events upon completion of field training. We study several versions of the algorithm, including constraints that encourage demographic diversity of FTOs. Using data from Indianapolis, we show that rewiring the network can increase the expected time (in days) of a recruit's first use of force incident by 8%. We then discuss the potential benefits and challenges associated with implementing such an algorithm in practice.

We review several concepts and modeling techniques from statistical and machine learning that have been developed to forecast recidivism. We show how these methods might be repurposed for forecasting police officer use of force. Using open Chicago police department use-of-force complaint data for illustration, we discuss feature engineering, construction of black-box models, interpretable forecasts, and fairness.

Interpretable models for criminal justice forecasting are desirable due to the high-stakes nature of the application. While interpretable models have been developed for individual level forecasts of recidivism, interpretable models are lacking for the application of space-time crime hotspot forecasting. Here we introduce an interpretable Hawkes process model of crime that allows forecasts to capture near-repeat effects and spatial heterogeneity while being consumable in the form of easy-to-read score cards. For this purpose we employ penalized likelihood estimation of the point process with a total-variation regularization that enforces the triggering kernel to be piece-wise constant. We derive an efficient expectation-maximization algorithm coupled with forward backward splitting for the TV constraint to estimate the model. We apply our methodology to synthetic data and space-time crime data from Indianapolis. The TV-Hawkes process achieves similar accuracy to standard Hawkes process models of crime while increasing interpretability and transparency.

Analysis and modeling of crime text report data has important applications, including refinement of crime classifications, clustering of documents, and feature extraction for spatio-temporal forecasts. Having better neural network representations of crime text data may facilitate all of these tasks. This paper evaluates the ability of generative adversarial network models to represent crime text data and generate realistic crime reports. We compare four state of the art GAN algorithms in terms of quantitative metrics such as coherence, embedding similarity, negative log-likelihood, and qualitatively based on inspection of generated text. We discuss current challenges with crime text representation and directions for future research.

Racial bias of predictive policing algorithms has been the focus of recent research and, in the case of Hawkes processes, feedback loops are possible where biased arrests are amplified through self-excitation, leading to hotspot formation and further arrests of minority populations. In this article we develop a penalized likelihood approach for introducing fairness into point process models of crime. In particular, we add a penalty term to the likelihood function that encourages the amount of police patrol received by each of several demographic groups to be proportional to the representation of that group in the total population. We apply our model to historical crime incident data in Indianapolis and measure the fairness and accuracy of the two approaches across several crime categories. We show that fairness can be introduced into point process models of crime so that patrol levels proportionally match demographics, though at a cost of reduced accuracy of the algorithms.