An Application of Machine Learning for Predicting Rearrests: Significant Predictors for Juveniles

This study examined the influence of the individual and social background in predicting the rearrest of 1124 juveniles who were first brought into the juvenile detention center in a midsized county in California. Independent variables include demographic characteristics, first offense type, gang affiliation, usage of drug, and family characteristics. Using cross-validation to choose an appropriate machine learning model for predicting rearrest, this study identified that the most important predictors of subsequent arrest are age at first arrest, drug usage, gang affiliation, and family with government assistance. Despite the fact that blacks are overrepresented in the juvenile detention population, race was not a significant predictor for rearrest. Future research would continue to explore the utilization of machine learning adding nontraditional variables to enhance the prediction of recidivism.