AI-Driven Exercise Prescription and Personalized Training: Adaptive Algorithms in Digital Health and Telerehabilitation

Abstract

Introduction and Objective. Artificial intelligence (AI) and machine learning (ML) are increasingly being 
applied to exercise prescription, personalized training, digital health and telerehabilitation. Conventional 
exercise prescription generally relies on standardized guidelines and periodic assessments, whereas 
AI-driven systems can continuously integrate information about exercise behavior, physiological responses, 
performance, recovery and adherence to modify training recommendations. This narrative review examines 
the potential of adaptive algorithms to support individualized exercise prescription and remote 
rehabilitation. Material and method. The review considered applications of ML in exercise monitoring, 
personalized training, wearable technology and telerehabilitation, with particular attention to physiological, 
behavioral and performance data. Classification, regression, recommendation and longitudinal learning 
approaches were considered together with data quality, generalizability, explainability, privacy and clinical 
oversight. Results. Wearable-based ML systems can identify physical activity, estimate physiological 
responses and provide individualized exercise recommendations. Adaptive systems may modify exercise 
intensity, duration, frequency and progression according to changes in individual response. These 
approaches are particularly relevant to telerehabilitation because continuous monitoring can extend 
assessment beyond conventional clinical appointments. However, current evidence remains limited by small 
and homogeneous datasets, inconsistent validation, sensor error, limited longitudinal evidence and 
uncertainty regarding clinical effectiveness. Conclusions. AI provides a promising decision-support 
framework for personalized exercise prescription and telerehabilitation. Current evidence supports AI 
primarily as an augmentation of exercise professionals and rehabilitation clinicians rather than as an 
autonomous replacement. Future research should prioritize longitudinal clinical validation, individualized 
modelling, explainability, data governance and real-world implementation.