| The criminal justice system has a long history of attempting to predict outcomes for several criminal justice decision points. Failure to appear in court is one such area of research. The current study compares the accuracy of deductive and inductive statistical models in predicting failure to appear in court. To evaluate this question, logistic regression was compared to random forests, support vector machines, and naive Bayes models. The efficacy of a stacked ensemble model, a model developed from the predicted probabilities of the aforementioned individual models, was compared as well. Model accuracy assessment was determined using an identical holdout set of cases across all models. The results indicate that the random forest model outperforms logistic regression at both overall accuracy and a one percent false positive threshold. This adds to a growing body of literature that evaluates the efficacy of inductive models of prediction in criminal justice applications. Future research should continue to evaluate the efficacy of inductive and deductive statistical models in various criminal justice applications. Indeed, despite the performance of the random forest model presented here, many statistical models should be considered whenever any new prediction model is developed. Each model is developed in a unique way, with specific strengths and weaknesses, and only when several are considered is it possible to identify the best for a specific application. Since even modest increases in predictive accuracy can improve the efficiency and outcome of the criminal justice system, the search for the accurate prediction should continue. |
