Ensemble learning approach for delinquency risk assessment in adolescents

Abstract

Adolescent delinquency is defined as an adolescent or teenager’s tendency of committing criminal acts. Children develop more advanced reasoning patterns during the onset of adolescence (10-12 years age). Hence, there is a higher risk of development of deviance behavior in them during this period which may lead to involvement in delinquent activities in future. So, early detection of such behavior in adolescents is crucial. Due to the necessity to correlate a significant number of variables before making a final decision, human screening of these individuals by a psychologist is tough. As a result, some effective computer- assisted mechanism, such as machine learning, must be devised for the early detection of adolescents with suspicious deviant conduct utilizing quantitative analysis. Taking this problem into consideration an effort has been made to implement multiple machine learning based ensemble technique using bagged decision tree, stacked ensemble model and adaptive boosting for classification of delinquent behavior in adolescents into three groups viz. mild, moderate and severe based on the level of severity. The Indian version of the International Self-Report Delinquency Study (ISRD-3) questionnaire, along with the required regional rationalization has been used to analyze behavioral data of 126 adolescents from ten schools of Ranchi, Jharkhand, India. The simulation results on our dataset shows that the bagged ensemble model outperforms individual classifiers as well as stacked ensemble and AdaBoost meta classifiers for classification of juvenile delinquency in terms of accuracy, AUC (Area Under the Curve), Kappa Value and FI Score.