Dataset Information

Optimizing predictive performance of criminal recidivism models using registration data with binary and survival outcomes.

ABSTRACT: In a recidivism prediction context, there is no consensus on which modeling strategy should be followed for obtaining an optimal prediction model. In previous papers, a range of statistical and machine learning techniques were benchmarked on recidivism data with a binary outcome. However, two important tree ensemble methods, namely gradient boosting and random forests were not extensively evaluated. In this paper, we further explore the modeling potential of these techniques in the binary outcome criminal prediction context. Additionally, we explore the predictive potential of classical statistical and machine learning methods for censored time-to-event data. A range of statistical manually specified statistical and (semi-)automatic machine learning models is fitted on Dutch recidivism data, both for the binary outcome case and censored outcome case. To enhance generalizability of results, the same models are applied to two historical American data sets, the North Carolina prison data. For all datasets, (semi-) automatic modeling in the binary case seems to provide no improvement over an appropriately manually specified traditional statistical model. There is however evidence of slightly improved performance of gradient boosting in survival data. Results on the reconviction data from two sources suggest that both statistical and machine learning should be tried out for obtaining an optimal model. Even if a flexible black-box model does not improve upon the predictions of a manually specified model, it can serve as a test whether important interactions are missing or other misspecification of the model are present and can thus provide more security in the modeling process.

SUBMITTER: Tollenaar N

PROVIDER: S-EPMC6407787 | biostudies-literature | 2019

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Optimizing predictive performance of criminal recidivism models using registration data with binary and survival outcomes.

Tollenaar Nikolaj N van der Heijden Peter G M PGM

PloS one 20190308 3

In a recidivism prediction context, there is no consensus on which modeling strategy should be followed for obtaining an optimal prediction model. In previous papers, a range of statistical and machine learning techniques were benchmarked on recidivism data with a binary outcome. However, two important tree ensemble methods, namely gradient boosting and random forests were not extensively evaluated. In this paper, we further explore the modeling potential of these techniques in the binary outcom ...[more]

PMID: 30849094

Similar Datasets

Project description:BackgroundThe duration of forensic psychiatric care is in Sweden not determined at the time of sentencing; instead, offenders are regularly evaluated, often with regard to risk of criminal recidivism. The length and justifiability of such a sanction have been greatly debated; however, previous estimates of treatment duration based on datasets delimited to discharged patients-have provided an uncertain groundwork for these deliberations. The aim of this study was to use a more suitable approach to calculate average duration of forensic psychiatric care and to examine the relationship between length of treatment and subsequent recidivism after discharge.MethodsThis retrospective cohort study focused on offenders sentenced to forensic psychiatric care in Sweden between 2009 and 2019 and registered in the Swedish National Forensic Psychiatric Register (n = 2064), with a follow-up period until May 2020. We used Kaplan-Meier estimator to calculate and visualize treatment duration including analyses comparing levels of relevant variables, and then evaluated criminal recidivism in patients discharged from treatment between 2009 and 2019 (n = 640), after stratification for the same variables and dichotomization by treatment duration.ResultsThe median duration of forensic psychiatric care was estimated to 89.7 months (95% CI 83.2-95.8). Treatment was longer in offenders who committed violent crimes, suffered from psychosis, or had a history of substance use disorder, and in offenders whose sentences included special court supervision. The cumulative incidence of recidivism in patients discharged from treatment was estimated to 13.5% at 12 months (95% CI 10.6-16.2) and 19.5% at 24 months (95% CI 16.0-22.8). Corresponding cumulative incidence of violent crime post discharge was 6.3% at 12 months (95% CI 4.3-8.3) and 9.9% at 24 months (95% CI 7.3-12.4). Among other findings, in patients without a history of substance use disorder and patients whose sentences did not include special court supervision, recidivism was significantly higher in those with a shorter treatment duration.ConclusionUsing the entirety of a suitable, contemporary, prospectively enrolled cohort of mentally ill offenders, we were able to estimate-with greater accuracy than previous studies-the average duration of Swedish forensic psychiatric care and rate of subsequent criminal recidivism.

Dataset Information

Optimizing predictive performance of criminal recidivism models using registration data with binary and survival outcomes.

Publications

Optimizing predictive performance of criminal recidivism models using registration data with binary and survival outcomes.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets