Learnijoy
ResearchAI ResearchAI Engineering & DevTools

AI Models Predict Adolescent Substance Use Onset with High Accuracy

Yixuan He, Jinni Su, Yun Kang· August 18, 2026 View original

Key takeaways

  • Longitudinal data significantly improves substance use prediction in adolescents.
  • Combined temporal and graph-based AI models offer superior predictive accuracy.
  • Peer deviance, age, and externalizing symptoms are strong predictors.
  • Early identification allows for more targeted prevention strategies.

Who benefits

HealthcarePublic HealthSocial ServicesEducation

Summary

Researchers developed AI models, including temporal XGBoost and Temporal Graph Convolutional Networks, to predict adolescent substance use onset using longitudinal and graph-augmented data from the ABCD Study. Combining these approaches achieved AUC-ROC values above 0.79, identifying key predictors like peer deviance and externalizing symptoms.

Identifying adolescents at risk of substance use early is a critical public health challenge. This research explores the effectiveness of various AI modeling approaches, including cross-sectional, longitudinal, and graph-based methods, to predict the onset of alcohol and marijuana use. The study utilized extensive data from approximately 11,860 participants in the Adolescent Brain Cognitive Development (ABCD) Study. The findings indicate that models incorporating longitudinal data consistently outperform those relying solely on baseline characteristics. Specifically, temporal XGBoost showed strong standalone performance. While Temporal Graph Convolutional Networks (T-GCNs) did not surpass temporal XGBoost on their own, their graph-derived risk scores provided valuable complementary information. The most accurate predictions were achieved by combining temporal XGBoost and T-GCNs, reaching AUC-ROC values over 0.79. Key predictive factors identified include peer influence, age, externalizing behaviors, parental monitoring, cultural norms, and neighborhood context.

Why it matters

Professionals in public health, social services, and healthcare can leverage these advanced predictive models to identify at-risk adolescents earlier, enabling more targeted and effective prevention and intervention strategies.

How to implement this in your domain

  1. 1Explore integrating longitudinal data collection into existing risk assessment protocols.
  2. 2Pilot advanced machine learning models (e.g., temporal XGBoost) for risk prediction.
  3. 3Consider incorporating social network data to build graph-based risk signals.
  4. 4Develop early intervention programs tailored to the identified high-risk factors.
  5. 5Collaborate with data scientists to build and validate robust predictive tools.

Original post by Yixuan He, Jinni Su, Yun Kang

"arXiv:2608.14578v1 Announce Type: new Abstract: Early identification of adolescent substance-use risk is an important prevention challenge, yet the relative value of baseline characteristics, longitudinal trajectories, and relational context remains unclear. Using data from appro…"

View on X

Originally posted by Yixuan He, Jinni Su, Yun Kang on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI ResearchAI Engineering & DevTools

Digital Twin Simulates Liver Health and Disease Progression

Researchers developed HEPATWIN, a physiology-informed digital twin of the human liver that integrates metabolic processes and patient-specific inputs to simulate liver function and early-stage disease progression, generating clinically observable biomarker trajectories.

Sumaiya Afroz Mila, Sandip RayAug 18, 2026
AI ResearchAI Engineering & DevTools

Explaining Multi-Objective Reinforcement Learning with Counterfactuals

This paper introduces command-space counterfactual explanations for Pareto-Conditioned Networks (PCNs), allowing users to understand how slight shifts in desired return commands would alter an agent's actions in multi-objective reinforcement learning scenarios.

Joanikij Chulev, Hendrik BaierAug 18, 2026
AI Engineering & DevToolsAI Research

LLM Framework Generates and Verifies Parallel DEVS Statecharts

This research introduces PDEVS-LLM, an agentic framework that uses large language models to assist human modelers in generating and verifying Parallel Discrete Event System Specification (PDEVS) statecharts, improving accuracy through controlled correction and logical consistency checks.

Vamsi Krishna Vasa, Hessam S. Sarjoughian, Edward J. YelligAug 18, 2026