The aim of this systematic review was to evaluate the validity and accuracy of wearable technologies used for monitoring physiological metrics in team-sport athletes. A systematic literature search was conducted in PubMed and Scopus databases, with additional studies identified through supplementary searching. Studies published between 2015 and 2025 were included if they assessed wearable devices in team-sport populations and compared their measurements with gold-standard methods. A total of eleven studies met the inclusion criteria. The findings indicate that heart rate monitoring demonstrates consistently high validity across different wearable devices, particularly in controlled laboratory conditions. In contrast, energy expenditure estimation shows substantial variability and systematic underestimation, especially during high-intensity and intermittent activities typical of team sports. VO₂max estimation presents mixed validity depending on device type and testing protocol, while respiratory frequency measurement demonstrates high agreement with gold-standard methods when assessed using specialized devices. Overall, wearable technologies provide valuable insights into athlete monitoring; however, their accuracy varies considerably depending on the physiological parameter and testing environment. These findings highlight the need for improved validation protocols and caution in the application of wearable-derived data in high-performance team-sport settings.

Background: Artificial intelligence (AI) is increasingly applied to physiological prediction, physical activity monitoring, and personalized exercise prescription. However, evidence regarding its practical effectiveness remains heterogeneous. This systematic review examined applications of AI, machine learning (ML), and deep learning (DL) in exercise-related prediction, monitoring, and personalization among adults. Methods: Following PRISMA 2020 guidelines, PubMed and Scopus were searched for studies published between 2015 and 2025, with Google Scholar used as a supplementary source. Eligibility followed PICO: adults aged 18 years or older (P); AI, ML, or DL applications for exercise prediction, monitoring, or personalization (I); conventional or non-personalized approaches, where applicable (C); and predictive accuracy and physiological, functional, or behavioral exercise outcomes (O). Findings were synthesized descriptively. Results: Ten studies were included: four examined physiological or energy-expenditure prediction, two investigated activity-related behavioral outcomes, two evaluated training-response classification or recommendation generation, and two assessed AI-personalized exercise interventions. Predictive models demonstrated promising performance in specific applications, although validation procedures and populations varied substantially. The intervention studies reported greater perceived exercise intensity and satisfaction in one trial and improvements in selected measures of balance, flexibility, and strength in another. Conclusions: AI-based approaches show potential for supporting exercise assessment, monitoring, and individualized decision-making. However, evidence regarding their superiority over conventional exercise prescription remains limited by the small number of intervention studies, methodological heterogeneity, and insufficient external validation. Further well-designed, long-term intervention studies are needed to establish their comparative effectiveness and practical benefits.