Upon release from a psychiatric facility, 1 in 8 men and 1 in 16 women will commit a serious crime. Moreover, 1 in 20 psychiatric patients will commit a violent crime.
This study showed that predictive machine learning models have the ability to personalize the identification of psychiatric patients (rather than broad group characteristics) who are at risk for committing criminal offenses and exhibiting violent behavior.
The literature was searched for studies that used predictive machine learning models to detect psychiatric patients at risk for criminal and violent behavior. The models were developed using single-nucleotide polymorphisms, text analysis, psychometric scales, hospital records, and resting state regional cerebral blood flow.
The following measures were calculated based on nine predictive machine learning models in the literature: area under the curve, 0.816; sensitivity, 73.33%; specificity, 72.90%; and accuracy, 71.45%.