Game Log Data from Adolescent Substance Use Risk Study

160 adolescents aged 11–14 played an HIV prevention game, generating 240 gameplay-derived behavioral metrics. Kammarauche Aneni analyzed this data in a proof-of-concept study to predict lifetime substance use and drug-refusal self-efficacy using machine learning models. The dataset, last updated in 2026, contains the results of this analysis.

Use Cases

• Training predictive models for adolescent substance use risk based on in-game behavioral metrics.
• Benchmarking machine learning models (SVM, Logistic Regression, Gradient Boosting) on behavioral data.
• Studying the relationship between executive function, decision-making, and inhibitory control metrics and substance use.
• Investigating the feasibility of using digital games as a low-burden screening tool in clinical settings.

Strengths

• Contains 240 distinct gameplay-derived behavioral metrics per participant.
• Data originates from a randomized controlled trial with 333 participants, from which a subset of 160 adolescents was analyzed.
• Model performance is reported with specific metrics including AUC, sensitivity, and F-1 scores.

Limitations

• Predictive accuracy was insufficient for clinical translation, with mean AUC reported below 0.6 across all models.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

figshare

Collection Method

Game log data extracted from an HIV prevention game played by adolescents in a school-based randomized controlled trial.

Freshness

Last updated 2026-05-01 05:27:41; freshness should be verified.