Table 1.
Crime recidivism prediction: Rank of different models.
Fig 1.
Prevalence and model evaluation: TPR, TNR, PPV, NPV.
Fig 2.
Prevalence and model evaluation: F-scores, Falkes-Mallows and Jaccard.
Fig 3.
Prevalence and model evaluation: Accuracy, MCC, Gmean, AUC, BA, BI, Kappa.
Fig 4.
Model comparison: TPR, TNR, PPV, NPV, F-scores, Falkes-Mallows and Jaccard.
Fig 5.
Model comparison: Accuracy, MCC, Gmean, AUC, BA, BI, Kappa.
Fig 6.
Model rankings: TPR, TNR, PPV, NPV, F-scores, Falkes-Mallows and Jaccard.
Fig 7.
Model rankings: Accuracy, MCC, Gmean, AUC, BA, BI, Kappa.
Table 2.
Variance of rankings of models by different metrics.
Fig 8.
Model evaluation with different thresholds: TPR, TNR, PPV, NPV, accuracy, MCC, BA, BI, F1 score.
Fig 9.
Model evaluation with different thresholds: AUC, Falkes-Mallows, Jaccard, Gmean, Kappa, F-beta scores, markedness and DOR.
Fig 10.
Variance of evaluation metrics by number of thresholds.
Fig 11.
OLS results of number of thresholds on variance of evaluation metrics with 95 percent confidence intervals.
Table 3.
Crime recidivism model evaluation metric values.