Pathways to juvenile hacking: A stacking ensemble learning and decision tree analysis of cumulative risk factors

Purpose: While prior research has identified correlates of juvenile hacking, the cumulative and sequential pathways through which these factors interact remain underexplored. This study develops and validates a decision support system (DSS) to model the hierarchical pathways leading to juvenile hacking, with the aim of informing targeted prevention and intervention strategies. Methods: Utilizing data from the Second International Self-Report Delinquency Study (ISRD-2; N = 68,507 adolescents across 31 countries), a stacking ensemble learning (SEL) model combining 20 machine learning algorithms was developed to predict juvenile hacking. A decision tree interpretation model (DTIM) was then constructed to visualize the hierarchical interactions among key risk factors identified through SEL feature ranking. Results: The SEL model achieved high predictive accuracy (95.4%), demonstrating the efficacy of machine learning techniques for identifying at-risk youth. The DTIM revealed digital piracy as the strongest differentiator, serving as the root node in the decision pathways. Distinct etiological routes emerged, moderated by gender, substance use, family bonds, and technology access. The highest-risk subgroup (97.2% probability) comprised male digital pirates with high substance use, pro-violence attitudes, and mobile phone ownership. Conclusions: Juvenile hacking results from cumulative, multi-domain risk interactions rather than single causes. These findings provide an empirical foundation for targeted, multi-modal interventions within the juvenile justice system, enabling efficient resource allocation and informing prevention programs addressing the confluence of digital behavior, substance use, and family dynamics. The proposed DSS offers a scalable tool for enhancing cybersecurity while reducing investigative costs.