The Role of AI in Physical Activity Research Addressing Mental Health in Adolescents and Young Adults: A Scoping Review

Background
Mental health challenges, including depression and anxiety, are increasingly concerning for adolescents and young adults. While physical activity (PA) is well-documented to support mental well-being, few interventions employ personalized approaches. Artificial intelligence (AI)-driven programs offer promising solutions by enabling data-driven customization of PA. However, current research on AI-integrated PA interventions remains early and lacks comprehensive synthesis. This study conducted a scoping review to examine the role of AI in PA interventions and their key outcomes and identify which study designs have been employed in this field. It is guided by three primary research questions: (1) Which study designs are used to implement AI-driven PA interventions? (2) How is AI utilized in PA interventions addressing mental health among adolescents and young adults (3) What are the outcomes of the AI-driven PA interventions?

Methodology
We conducted a scoping review following the PRISMA-ScR guidelines. Given that 2015 marks the rise of deep learning–based AI applications, we searched for peer-reviewed studies published between 2015 and 2024 across eight databases. A total of 405 records were identified, and 151 duplicates were removed. After screening titles, abstracts, and full texts, 27 studies met the inclusion criteria. Eligible studies focused on individuals aged 15–29, implemented AI-driven PA programs, and reported mental health outcomes such as depression, anxiety, or stress. The screening process was facilitated using the Rayyan software.

Results
The 27 studies included 21 observational and 6 quasi-experimental studies. Study designs included wearable-based longitudinal observation, randomized controlled trials (RCTs), and crossover interventions using AI-driven messaging or personalized activity prompts. Most studies targeted adolescents and young adults aged 15 to 29. AI technologies were utilized in four primary ways: (1) behavioral data collection and prediction using wearable sensors and machine learning models (e.g., Random Forest, CNN), (2) personalized interventions via reinforcement learning-based messaging, (3) immersive activity programs using augmented reality, and (4) multimodal emotion prediction models combining gait, facial expressions, or EEG. Observational findings consistently indicated that reduced PA was associated with worsened mental health outcomes. A 10% decrease in daily step count predicted a 1.2-point increase in depression scores (p < .01), while skipping 24 minutes of moderate-to-vigorous PA predicted a 15% rise in stress. AI-based interventions demonstrated significant improvements, including a 1,250-step increase in daily activity and a 2.3-point reduction in depression (p < .01). CNN-based motion recognition achieved 92.98% accuracy, and multimodal models reached up to 82.2% in classifying depression symptoms. Personalized models outperformed group-based approaches in predictive accuracy.

Conclusion
AI-driven PA interventions offer a promising alternative to traditional group-based programs by enabling real-time monitoring and personalized feedback for adolescents and young adults. This review confirms that AI can accurately analyze and predict PA and its interactions with lifestyle factors such as sleep, diet, and emotional responses. Future research should address data privacy, algorithmic bias, and cultural adaptability while evaluating long-term impacts and ethical scalability. Interdisciplinary collaboration will be essential for integrating AI into public health and education policy.