Classification of Juvenile Delinquency Using Bayesian Network Learning: A Comparative Analysis

Abstract

The practice of engaging in offensive behavior on a frequent basis by a teenager is referred to as juvenile delinquency. Data mining and machine learning have been very effective techniques for a long time, allowing for efficient and accurate prediction in a variety of real-world applications. These techniques are gradually being implemented internationally in the area of criminal behavioral analysis, particularly in the detection of adolescent delinquency. According to studies, the risk of developing a deviant personality rises exponentially throughout the early period of adolescence. As a result, it makes perfect sense to identify deviant teenagers early and provide appropriate medical counselling. Providing routine psychological screening services for teenagers in a densely populated country is exceedingly difficult. Furthermore, due to a dearth of skilled clinicians, human evaluation of individual teenage behavior is highly subjective and time consuming. To handle this problem, an automated framework for the early identification of delinquent activity in juveniles has been implemented using Bayesian Network learning techniques. In this research, multi-class classification has been carried out using multiple Bayesian network based learning algorithms viz. K2 search, Simulated annealing, LAGD Hill Climbing and Tabu search. 5-Fold Cross-validation has been employed for multi class classification of juveniles into three groups based on severity levels viz. low, moderate and extreme. Simulation results are obtained and a comparative analysis shows that Bayesian Network with LAGD Hill Climber outperforms all other techniques.