Predicting Recidivism to Drug Distribution using Machine Learning Techniques

Abstract

Recidivism is an important issue in imprisonment and probation processes. The aims of this research are to find crucial factors for predicting recidivism to drug distribution and to investigate the power of machine learning for the recidivism prediction. Our proposed approach employed a data set containing 598 inmates to establish and evaluate a feature selection algorithm and machine learning-based recidivism models. The experimental results show that almost recidivism prediction models with selected factors perform better than or equal to the models with all factors. Additionally, the results point out top four important factors composed of royal pardons or suspension, first offending age, encouragement of family members, and frequency of substance use. This concludes that machine learning techniques with the help of a feature selection algorithm can be a promising approach for the recidivism prediction in which the government can exploit to find a suitable prison rehabilitation program.