Table 2 Prediction results across machine learning models.

From: Machine learning of clinical and neural data predicts future homicide in high-risk youth

Model

Model performance (%)

overall/sensitivity/specificity

ROC AUC (%)

A priori model with neural data + Age + PCL: YV scores

76/86/74

82

Clinical data only

68/14/80

65

Neural data only

71/71/71

80

Clinical + neural data

68/86/65

80

  1. Numbers in model performance column represent percentages of overall accuracy, sensitivity (i.e., true positive), and specificity (i.e., true negative) rates, respectively. Area under the receiver operating characteristic (ROC) curve results are presented in the last column. The ROC curve plots true positive rates against false positive rates for all decision thresholds. The ROC AUC computes the area under the ROC curve, with a higher value indicating better ability to predict homicide group members (positive) while minimizing false positives (1: perfect predictive model, 0.5: random guess model). A priori model is listed first. Remaining models were run for completeness. Clinical Data = Age, IQ, socioeconomic status (SES), number of substance use disorders (SUD), addiction severity index (ASI) substance use severity scores, Barrett impulsiveness scale (BIS-11) total scores, psychopathy checklist: youth version (PCL: YV) factor 1 and factor 2 scores, number of traumatic brain injuries (TBI), age of first arrest, trauma checklist 2.0 (TCL) factor 1 and factor 2 scores, number of prior felonies, probation/parole violations, and convictions, psychopathology assessed via kiddie schedule for affective disorders and schizophrenia (KSADS) (i.e., anxiety, depression, conduct disorder/ oppositional deviant disorder (CD/ODD), attention deficit/ hyperactivity disorder (ADHD), post-traumatic stress disorder (PTSD)), gang affiliation, parental incarceration, parental separation; neural Data = GMV within a priori ROIs including the left and right amygdala, left and right insula, left and right parahippocampal gyrus, left and right middle temporal pole, left and right superior temporal pole, left and right orbitofrontal cortex, and total brain volume (BV).
Back to article page