Title : PIVOT: Predictive intervention for vaping using resonance modeling of oscillatory, psychological, and trigger dynamics
Abstract:
Vaping affects over 100 million people worldwide and is the leading form of tobacco use among youth. Yet existing interventions remain reactive, non-individualized, and achieve only 24–50% short-term cessation success. The central scientific gap is temporal: no current framework identifies when a specific individual enters a window of heightened vaping vulnerability before craving onset. This paper presents PIVOT, grounded in Addiction Resonance Theory (ART), a novel framework modeling vaping vulnerability as resonance arising from three interacting systems: a Biological Layer of physiological oscillators, a Bee Layer of environmental trigger modulation, and an Octopus Layer of competing neural decision modules. PIVOT employs a three-phase physiologically constrained digital data amplification pipeline: a real survey of 973 adult vapers, calibrated stochastic expansion to 10,000 digitally amplified profiles (24M raw data points), and lossless Nyquist-Shannon compression to 8.65M training samples (63.9% volume reduction). Nine ML models are evaluated; the Neural Network achieves the highest accuracy (94.1%, AUC 0.988) while CatBoost is selected as the production model (93.4%, AUC 0.986) for inference speed, deployment simplicity, and exact SHAP interpretability via TreeExplainer. Tenfold cross-validation confirms 93.39% ± 0.15% accuracy. Layer-wise ablation validates ART: no single layer exceeds 74.6% accuracy; only their combination reaches 93.4%. SHAP analysis identifies stress, social proximity, and goal-setting as dominant predictors of vaping vulnerability, consistent with findings reported by Rahimi et al. across an independent cohort of 1,119 vapers.
