Interpretable classification models for recidivism prediction

By Jiaming Zeng, Berk Ustun, Cynthia Rudin. 2015

270 citations recorded by the source corpus.

We investigate a long‐debated question, which is how to create predictive models of recidivism that are sufficiently accurate, transparent and interpretable to use for decision making. This question is complicated as these models are used to support different decisions, from sentencing, to determining release on probation to allocating preventative social services. Each case might have an objective other than classification accuracy, such as a desired true positive rate TPR or false positive rate FPR. Each (TPR, FPR) pair is a point on the receiver operator characteristic (ROC) curve. We use popular machine learning methods to create models along the full ROC curve on a wide range of recidivism prediction problems. We show that many methods (support vector machines, stochastic gradient boosting and ridge regression) produce equally accurate models along the full ROC curve. However, methods that are designed for interpretability (classification and regression trees and C5.0) cannot be tuned to produce models that are accurate and/or interpretable. To handle this shortcoming, we use a recent method called supersparse linear integer models to produce accurate, transparent and interpretable scoring systems along the full ROC curve. These scoring systems can be used for decision making for many different use cases, since they are just as accurate as the most powerful black box machine learning models for many applications, but completely transparent, and highly interpretable.

Subfields: Artificial Intelligence, Safety Research, Safety, Robustness & Interpretability, AI Ethics, Fairness & Society

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Safety Research

In Safety, Robustness & Interpretability

In AI Ethics, Fairness & Society

Citations in the corpus

The accuracy, fairness, and limits of predicting recidivism (2018, positive): “This finding is supported by earlier work that showed how to create simple, transparent, and interpretable predictive rules for recidivism prediction ( 7 , 14 , 15 ).”
Fair-by-design explainable models for prediction of recidivism (2019, background): “Therefore, it is highly important to guarantee that such instruments are free from discriminatory biases [5].”
Please Stop Explaining Black Box Models for High Stakes Decisions (2018, background): “Figure 4: Scoring system for risk of recidivism from Rudin and Ustun [2018] [which grew out of Ustun and Rudin, 2017, Zeng et al., 2017, Ustun and Rudin, 2015].”
Interventions over Predictions: Reframing the Ethical Debate for Actuarial Risk Assessment (2017, background): “In addition to transparency issues related to proprietary software, a major focus of recent scholarship has been on developing interpretable machine learning models that enable outside researchers to better scrutinize the underlying logic of actuarial tools which are based on machine learning (Zeng ”
Auditing black-box models for indirect influence (2016, methodology): “For this work, we processed the data additionally following the description of the processing performed in a previous study on interpretable models [29].”
Path-Specific Counterfactual Fairness (2018, background): “The simplest solution to this challenge is to down-weight or discard the sensitive attribute (Zeng et al., 2016).”
Learning Certifiably Optimal Rule Lists (2017, neutral): “Similar observations have been reported for other datasets, namely that complex machine learning models do not have an advantage over simpler transparent models (Tollenaar and van der Heijden, 2013; Bushway, 2013; Zeng et al., 2017).”
Why Are We Using Black Box Models in AI When We Don’t Need To? A Lesson From An Explainable AI Competition (2019, methodology): “…methods applied to many different types of rearrest prediction problems on other datasets: the interpretable models (which were very small linear models or logical models in these studies) performed just as well as the more complicated (black box) machine learning models (Zeng et al., 2016).”
Learning Fair Rule Lists (2019, background): “For instance, when they have small or of reasonable size [44], ∗ Principal contributor. the following models can be considered as interpretable: rules sets [18, 43, 46, 54], rule lists [3, 59, 61], decision trees [9, 51] and scoring systems [39, 57, 64].”
Learning customized and optimized lists of rules with mathematical programming (2018, background): “0 often tended to produce trivial models that predict only one class, Zeng et al [75] provide experiments showing that C5.”
Learning Optimized Risk Scores (2016, methodology): “RiskSLIM risk scores are the risk assessment counterpart to SLIM scoring systems (Ustun et al., 2013; Ustun and Rudin, 2016; Zeng et al., 2016; Ustun et al., 2016).”
Explaining Explanations in AI (2018, background): “Historically, much work has focused on global approximations of models (e.g. Craven and Shavlik (1996); Martens et al. (2007); Sanchez et al. (2015), including approaches based on clustering (Chen et al., 2016), integer programming (Zeng et al., 2017), and rule lists (Wang and Rudin, 2015)).”
Optimal randomized classification trees (2021, neutral): “Extracting knowledge from data is a crucial task in Statistics and Machine Learning, which has applications in areas such as Biomedicine [1, 2], Criminal Justice [3, 4, 5], Fraud Detection [6], Privacy Protection [7], Health Care [8, 9, 10, 11], Risk Management [12, 13], Social Networks [14], and Co”