Utilizing machine learning algorithms for personalized workout recommendations and monitoring: A systematic review on smartwatch-assisted exercise prescription

Abstract

Background

Smartwatches, equipped with advanced sensors, have become increasingly prominent in health and fitness domains. Their integration with machine learning (ML) algorithms presents novel opportunities for personalized exercise prescription and physiological monitoring.

Objective

This systematic review aimed to evaluate the effectiveness, limitations, and practical applications of smartwatch-ML systems in delivering tailored fitness interventions and health tracking.

Methods

Following PRISMA guidelines, five databases (PubMed, Scopus, IEEE Xplore, Web of Science, and SPORTDiscus) were searched for studies published from January 2000 to December 2023. Inclusion criteria required empirical studies involving human participants, the use of smartwatches for exercise monitoring or prescription, and the application of ML algorithms. Forty-nine studies met the eligibility criteria and were synthesized narratively using thematic clustering.

Results

The majority of included studies demonstrated high algorithmic performance in activity recognition (>98% accuracy) and vital sign tracking. However, external validity was often limited due to lab-based testing, narrow demographic representation, and lack of standardized evaluation frameworks. Few studies incorporated explainable artificial intelligence, behavioral adaptation, or longitudinal validation. Ethical and regulatory considerations were rarely addressed.

Conclusion

Smartwatch-ML integration holds substantial promise for individualized, real-time health support, especially in fitness and rehabilitation. To ensure broader impact and clinical adoption, future research must address generalizability, ethical data governance, interpretability, and interdisciplinary system design.

Keywords

Smartwatches exercise prescription machine learning personalized fitness physiological monitoring wearable health technology activity recognition

Introduction

In recent years, the fusion of wearable technology and artificial intelligence (AI) has sparked a quiet revolution in personal health and fitness. Among these wearables, smartwatches have emerged as powerful tools, not just for tracking steps or telling time, but for collecting rich physiological and behavioral data in real time. These devices, equipped with sensors for heart rate, motion, temperature, and even sweat analysis, are now being paired with machine learning (ML) algorithms to deliver more than just insights; they are beginning to offer personalized, adaptive workout plans tailored to individual users.

The need for such personalized fitness support is pressing. Traditional exercise prescriptions, often based on general population averages, don’t account for the wide genetic and physiological variability in how individuals respond to physical activity.¹ Advances in ML, particularly those dealing with large and complex datasets, now make it possible to recognize patterns in real-world physiological signals and user behaviors, helping turn smartwatches into real-time exercise advisors. Studies have already shown that smartwatch-based systems can classify daily activities with over 98% accuracy and track motion in clinical settings with high usability.^2,3 Clinical-grade heart rate and blood pressure tracking via smartwatches is no longer futuristic; it's a current reality.⁴

Despite these impressive advancements, major gaps remain. While machine learning offers incredible potential for health personalization, traditional algorithms often struggle to process high-volume, high-speed data such as real-time sensor streams.^5,6 Many systems still fall short when applied to diverse, real-world conditions, particularly when tested outside of controlled environments or homogeneous populations.⁷ There are also growing concerns around the ethical, privacy, and practical integration aspects of these technologies, especially in clinical and public health settings.^8,9

This review aims to explore the current landscape of smartwatch-assisted exercise prescription powered by machine learning. Specifically, we seek to understand how these technologies are being applied in fitness, rehabilitation, and healthcare; evaluate the outcomes they generate; and identify the critical research, technical, and ethical gaps that must be addressed. By organizing and analyzing empirical studies across domains, this paper sheds light on the strengths and limitations of current approaches and envisions the future of personalized, adaptive exercise support through wearable platforms.

Background and theoretical framework

Smartwatch technologies

Smartwatches have evolved from simple step counters to sophisticated health-monitoring systems packed with sensors capable of tracking various physiological and biomechanical signals. Common sensors include photoplethysmography (PPG) for heart rate, electrodermal activity (EDA) sensors for stress, inertial measurement units (IMUs) for motion tracking, GPS modules for location data, and, more recently, sweat sensors for hydration and biochemical analysis.^10–12 Electrochemical and optical biosensors enable continuous, real-time diagnostics, while newer models even monitor glucose and electrolyte levels through sweat.¹³ These devices collect large volumes of diverse data, including heart rate variability (HRV), sleep cycles, physical activity patterns, and blood pressure, making them valuable tools in both fitness and clinical environments.^14,15

ML in healthcare and fitness

The integration of ML with smartwatch data has opened new possibilities for predictive healthcare and personalized fitness. ML algorithms can classify physical activity, detect anomalies, and provide tailored feedback. Decision tree classifiers, for example, have achieved up to 99.63% accuracy in recognizing human activities from multi-sensor data.¹⁶ The Association of Information Brokers of Small and Medium-sized Businesses in the Netherlands framework (AIBSNF) combines biosensors, real-time location systems (RTLSs), and ML models to enable context-aware, data-driven interventions in healthcare and sports.¹⁷ However, managing the volume, variety, and velocity of wearable data presents challenges. While deep learning offers the potential for uncovering complex patterns in Big Data, it struggles with streaming and high-dimensional inputs common in real-time health monitoring.^18,19

Exercise prescription principles

Exercise prescription involves designing a workout plan based on individual goals, fitness levels, health conditions, and responses to previous training. Foundational principles include specificity, progression, overload, and reversibility. Effective exercise programs must balance intensity, duration, and frequency to improve adherence and outcomes.²⁰ Cardiovascular stress testing guidelines help define safe and effective thresholds for exertion, particularly in clinical populations.²¹ However, uniform prescriptions often fail to account for personal variability; what works for one person may not work for another.

The need for personalization in workouts

Personalization is increasingly recognized as crucial for effective and sustainable exercise routines. Genetic and physiological differences can significantly affect how individuals respond to the same workout.¹ Smartwatches, by continuously tracking physiological signals, provide an opportunity to tailor exercise plans in real time. Personalized feedback enhances motivation, adherence, and performance, especially when informed by machine learning models trained on the user's unique data. This tailored approach not only improves fitness outcomes but also supports long-term health and well-being.

Theoretical framework: The personalization-feedback loop

This study is grounded in the personalization-feedback loop framework, which synthesizes concepts from self-regulation theory, adaptive control systems, and personalized medicine. At its core, this framework posits that:

Personalized exercise prescriptions derived from real-time biometric feedback and machine learning form a dynamic feedback loop that improves engagement, optimizes performance, and supports sustained health outcomes.

This theoretical model incorporates three interdependent components:

Sensing and data acquisition: Smartwatches collect multimodal physiological and biomechanical data streams that reflect current fitness status and training response.

Adaptive feedback system: ML algorithms process these data to generate personalized recommendations, adjusting variables such as intensity, duration, and type of exercise in real time.

Behavioral response and outcome monitoring: Users respond to these recommendations, and their physiological responses feed back into the system, allowing further refinement of future prescriptions. This continuous loop embodies principles of behavioral reinforcement and self-optimization.

This framework aligns with precision health paradigms and cybernetic models of behavior change, supporting a data-informed, user-centered approach to fitness and clinical rehabilitation. It also provides a conceptual foundation for evaluating the efficacy, ethical implications, and technological demands of ML-integrated wearables.

Anchored by the personalization-feedback loop framework, the present review interrogates how effectively current smartwatch-based systems leverage these components to deliver personalized, adaptive, and evidence-based exercise interventions.

Methods

This systematic review was conducted following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. The aim was to identify and synthesize studies that investigate the integration of smartwatches and ML algorithms for personalized exercise prescription and monitoring. The process included clearly defined inclusion and exclusion criteria, a comprehensive search strategy across multiple databases, and a transparent screening and selection process.

Inclusion and exclusion criteria

To ensure relevance and quality, studies were included based on the following predefined criteria: (1) examined the use of smartwatches or comparable wearable devices for exercise prescription or fitness monitoring utilizing ML algorithms, (2) written in English, (3) involved human participants, (4) presented original empirical data, (5) reported quantifiable outcomes such as algorithm accuracy, exercise efficacy, or health-related results, and (6) contained sufficient methodological detail for quality appraisal. Studies retrieved through citation chaining were subjected to the same criteria.

Studies were excluded if they (1) were case reports or single-subject designs, (2) involved participants with significant comorbidities (e.g. cardiovascular disease, stroke, diabetes, and chronic kidney disease), or (3) lacked methodological transparency or quantifiable data. Conference abstracts and posters were only considered if a peer-reviewed, extended version was available. Table 1 shows the inclusion and exclusion criteria of the literature search.

Table 1.

Inclusion and exclusion criteria.

Criteria type	Description
Inclusion criteria	Use of smartwatches/wearables for exercise prescription with machine learning (ML); human studies; empirical data; English language; reported outcomes; and sufficient methodology
Exclusion criteria	Case studies; participants with chronic illnesses; non-peer-reviewed conference proceedings; and studies lacking quantifiable data

Search strategy

A systematic literature search was carried out across five databases: PubMed, Scopus, IEEE Xplore, Web of Science, and SPORTDiscus. The search covered publications from January 2000 to December 2023. The final search was completed on 10 January 2024.

The search strategy used combinations of Medical Subject Headings (MeSH) and free-text keywords relevant to the review topic. Boolean operators were applied to refine results. The core search terms included:

MeSH terms and keywords

“Smartwatch,” “Wearable technology,” “Exercise prescription,” “Physical activity monitoring,” “Machine learning,” “Artificial intelligence,” “Fitness tracking,” “Sensor data,” “Personalized exercise,” “Biometric feedback,” “Rehabilitation,” and “Health outcomes.”

All identified records were imported into EndNote for reference management. After removing duplicates, titles and abstracts were screened for relevance. Full texts of potentially relevant articles were reviewed against inclusion/exclusion criteria. Additionally, references to included articles were manually searched to identify any further eligible studies.

Extraction of study characteristics

Data from the final included studies were systematically extracted and charted using a predefined framework. The following variables were collected: publication year, sample size, type of wearable sensor(s), ML models utilized, data processing methodologies, evaluation metrics (e.g. accuracy, precision, and F1-score), and key findings. These data are summarized in the results section and tabulated for clarity.

Population, interest, and context (PICo) framework

Structured research question (PICo format)

How are smartwatches, integrated with ML algorithms, being utilized to personalize exercise prescription and monitoring in individuals across fitness and health-related contexts?.

Table 2 shows the PICo of the studies. This formulation supports a broad, exploratory review by focusing on how and in what ways technologies are being applied, rather than evaluating a narrowly defined intervention.

Table 2.

Population, interest, and context of the studies.

Element	Description
Population (P)	Individuals engaging in physical activity or exercise (general population, fitness enthusiasts, patients in rehabilitation, etc.)
Interest (I)	Use of smartwatches and wearable biosensors in combination with machine learning algorithms for personalized exercise prescription and monitoring
Context (Co)	Health, fitness, and rehabilitation settings where wearable technology is used for real-time physiological monitoring and tailored workout recommendations

Study selection flow

The study selection process is illustrated in Figure 1, which summarizes the number of records identified, screened, assessed for eligibility, and included. Specific reasons for excluding full-text articles during the eligibility phase are also indicated.

Figure 1.

PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flow diagram of the study selection process.

A total of 4212 studies were retrieved from the database search. After removing 423 duplicates and screening 3351 titles/abstracts for relevance, 438 full-text articles remained. Of the resulting full-text studies, 389 were excluded for the following reasons:

127 lacked empirical data

98 focused on unrelated outcomes

92 used non-wearable technology

72 had inadequate methodological transparency

The left 49 studies that met the full criteria and were included in the qualitative synthesis. A Grading of Recommendations, Assessment, Development and Evaluation (GRADE) assessment was performed to evaluate the certainty of evidence across key outcomes related to wearable technologies.

Data analysis and quality appraisal

A general assessment of methodological soundness and potential bias was considered during the study selection and synthesis process to ensure a balanced representation of findings. Each study was independently evaluated by two reviewers for methodological quality, and disagreements were resolved through discussion. Effect measures varied based on reported outcomes, with accuracy, precision, recall, and F1-score being the most commonly used metrics for assessing machine learning model performance, while mean differences were used in studies reporting physiological or fitness improvements. To synthesize results, we employed a narrative synthesis approach, given the heterogeneity in study designs, ML techniques, and outcome measures; however, thematic clustering was used to organize findings by device type, ML model, and application domain. The risk of bias due to missing results was evaluated using a visual assessment of publication patterns and cross-referencing with clinical trial registries where applicable. The overall certainty of the evidence was assessed using the GRADE approach, considering factors such as the risk of bias, inconsistency, indirectness, imprecision, and publication bias to provide a transparent summary of confidence in the synthesized findings.

Results

This systematic review synthesized findings from 60 peer-reviewed studies to evaluate the feasibility and current capabilities of smartwatch-assisted exercise prescription powered by ML. The studies encompassed a wide range of applications, from activity recognition and cardiovascular monitoring to real-time feedback systems and algorithmic personalization. The results are grouped into six major thematic categories. Table 3 shows the summary of the reviewed studies.

Table 3.

Summary of selected studies on smartwatches, wearables, and ML in exercise prescription and monitoring.

Author	Year	Device type	ML algorithm used	Sample size	Outcome measured	Key finding
Balu and P¹⁶	2022	Wearable multi-sensor devices	Decision tree, others	Not stated	Human activity recognition accuracy	Achieved 99.63% accuracy in HAR using sensor fusion and supervised ML algorithms
Antognoli et al.²²	2023	Commercial smartwatches	Not specified	Not stated	HR and BP measurement accuracy	HR bias at 0.2 bpm; BP within acceptable confidence intervals, comparable to medical devices
Phatak et al.¹⁷	2021	BSN, RTLS, Wearables	AI/ML (unspecified types)	Not applicable (framework review)	Data fusion for healthcare and sport	AIBSNF enables synchronized physiological and positional data for ML analysis
Auepanwiriyakul et al.²³	2020	Consumer smartwatches	Not specified	Not stated	Inpatient motion tracking accuracy	Smartwatches accurately monitor hospital patient motion and are acceptable to users
L’Heureux et al.²⁴	2017	Not device-specific (Big Data context)	SVM, classification algorithms	Not applicable	Big Data ML challenges and solutions	ML challenges in Big Data outlined; new techniques proposed for better algorithm design
Najafabadi et al.¹⁸	2015	Not device-specific	Deep learning	Not applicable	Big Data analytics in DL	Deep learning effective in pattern discovery; challenged by dimensionality and stream data
Gibson et al.²⁰	2019	Not wearable-focused	Not applicable	Not applicable	Fitness assessment and exercise prescription	Book provides assessment standards for fitness domains including flexibility and VO₂ max
Fletcher et al.²¹	2013	Not wearable-focused	Not applicable	Not applicable	Exercise stress testing for cardiovascular response	Defines standards and mechanisms of cardiovascular stress during exercise testing
Ross et al.¹	2019	Not wearable-focused	Not applicable	Not specified	Exercise response variability	Variability in cardiorespiratory response linked to genetics and inter-individual factors
Bayoumy et al.²⁵	2021	Smart wearables	Not specified	Not applicable	Cardiovascular health monitoring	Wearables offer promise but suffer from data quality variation and lack of integration
Kario²⁶	2020	Wearable BP Monitor	Not specified	Not specified	BP monitoring and prediction	Wearables allow stress-free BP monitoring; support anticipatory cardiovascular care
Dehghani²⁷	2018	Smartwatch	Not specified	Netnography (qualitative)	Usage intention	Eight key determinants affect ongoing smartwatch use
Kim and Shin²⁸	2015	Smartwatch	Not specified	Survey-based	Tech acceptance	AQ, RA impact usefulness; MB, AV influence ease of use
Majumder et al.²⁹	2017	Wearable sensors	Not specified	Not specified	Vital sign monitoring	Cost-effective elderly monitoring with smart textiles
Dias et al.³⁰	2018	Wearable health devices	Not specified	Not specified	Vital signs and system design	Cardiovascular WHDs reviewed; highlighted privacy and tech trends
Sharma et al.¹⁰	2021	Biosensors	Not specified	Not specified	Biomarker monitoring	Optical/electrochemical biosensors aid continuous health diagnostics
King and Sarrafzadeh³¹	2017	Smartwatch	Not specified	Review of studies	Chronic disease monitoring	Feasibility studies dominated; limited validation with clinical tools
Henriksen et al.³²	2018	Smartwatches, fitness trackers	Not specified	423 devices	Physical activity measurement	Fitbit, Garmin, and Apple most used in research from 2011 to 2017
Theurl et al.³³	2023	Smartwatch	Not specified	Clinical comparison	Heart rate variability	Smartwatch HRV aligns with ECG in global markers; caution in high-frequency HRV
Kristoffersson and Lindén⁷	2022	Wearable sensors	Not specified	Multiple studies	Physical activity monitoring	Sensor studies flawed in demographic representation and real-life testing
Spann⁹	2016	Wearable fitness devices	None	Not applicable	Privacy risks	Laws needed to protect wearable users’ personal health data (Spann, 2016)
King and Jessen³⁴	2014	Smart meters	None	Not Applicable	Data Privacy	Smart meter data lacks strong legal privacy controls (King and Jessen, 2014)
Anaya et al.⁸	2017	Wearable devices	None	Survey-based	Ethical perceptions	Users demand better informed consent and shorter privacy policies (Anaya et al., 2017)
Datta et al.³⁵	2018	Wearable devices	None	Literature review	Privacy challenges	Research gaps exist in consumer wearable device privacy (Datta et al., 2018)
Min et al.³⁶	2023	Sweat sensors	None	Review	Personalized health monitoring	Sweat sensors enable predictive and personalized healthcare (Min et al., 2023)
Guk et al.³⁷	2019	Wearable biosensors	None	Review	Disease monitoring	Wearables evolving toward continuous real-time monitoring (Guk et al., 2019)
Soh et al.³⁸	2015	Wearable wireless health monitoring	None	Review	System challenges	Wearable systems face hurdles in reliability, privacy, and energy efficiency (Soh et al., 2015)
Darwish and Hassanien³⁹	2011	Wearable and implantable sensors	None	Review	Healthcare monitoring	Sensor networks enhance healthcare but face design challenges (Darwish and Hassanien, 2011)
Patel et al.⁴⁰	2012	Wearable sensors	None	Review	Rehabilitation applications	Wearables support rehabilitation but need clinical deployment strategies (Patel et al., 2012)
Vijayan et al.⁴¹	2021	Wearable sensors	None	Review	Data collection	Massive data from wearables offers insights, but comes with collection challenges (Vijayan et al., 2021)
Ahmed et al.⁴²	2020	AI platforms (EHR-integrated)	Various ML methods	Not applicable	Precision medicine integration	AI-enhanced ML platforms optimize clinical decision-making for tailored treatments
Mannini and Sabatini⁴³	2010	On-body accelerometers	HMM, PCA, and ICA	Not specified	Physical activity classification	Sequential classifiers achieved high accuracy; subject-specific training needed
Zakeri et al.⁴⁴	2008	Heart rate and accelerometers	CSTS model	Children sample	Energy expenditure	CSTS model accurately estimated TEE in children using HR and PA variables
Kang and McAuley⁴⁵	2018	Recommender systems (not wearable)	SASRec	Not specified	Recommendation performance	SASRec outperformed CNN/RNN, especially on sparse data, with higher efficiency
Shiri et al.⁴⁶	2023	Not wearable-specific	CNN, RNN, LSTM, GRU	Three datasets	Deep learning performance	DL models compared across domains; real-world complexity remains a challenge
De Fazio et al.⁴⁷	2023	Wearable sensors	Not specified	Review	Sports and rehab monitoring	Wearables monitor physiological rehab/sports parameters, but are costly and clinical-focused
Niess et al.⁴⁸	2020	Fitness trackers	Not applicable	Experimental sample	Visual feedback and rumination	Bar graphs reduce rumination better than multicolored visuals in fitness tracking
Morris and Aguilera⁴⁹	2012	Mobile, social, and wearables	Not specified	Not applicable	Psychological interventions	Mobile tech reshapes therapy through self-tracking and social integration
Wang et al.⁵⁰	2021	Smart wearables	XGBoost and SVM	Not specified	Fitness behavior and weight loss	Wearables initially help but show reduced use; fractal weight patterns observed
Burns et al.⁵¹	2020	Smartwatch	ML (unspecified)	Longitudinal cohort	Exercise adherence in physiotherapy	Smartwatch validated for tracking shoulder exercise adherence
Ren et al.⁵²	2018	Fitness tracker	PCFT (cooperative model)	Office workers	Physical activity	PCFT improved physical activity and social engagement in offices
Ghanvatkar et al.⁵³	2018	Personalized PA tools	None specified	Literature review	Personalization strategies in PA interventions	Six personalization types identified; models rely on PA profiles and BCTs
Sullivan and Lachman⁵⁴	2017	Fitness technology	Behavior change techniques	Sedentary adults	Physical activity behavior	Fitness tech with social support improves PA in sedentary adults
Keogh et al.⁵⁵	2021	Wearable sensors	Not specified	Expert interviews	Utility of wearables in gait/activity measurement	Experts favor wearable tech over questionnaires for real-world insight
Banaee et al.⁵⁶	2013	Wearable sensors	Data mining algorithms	Literature review	Health monitoring systems	Data mining enables wearable-based health monitoring but faces hurdles
Netz et al.⁵⁷	2021	Smartphone sensors	Not specified	Elderly adults (pilot)	Fitness improvement	Personalized remote programs improved senior fitness via sensors
Sartori et al.⁵⁸	2016	Neuro-musculoskeletal tech	Neural-driven models	Not stated	Neurorehabilitation outcomes	Models connect neural signals to muscles for tailored rehab interventions
Popham et al.⁵⁹	2023	Wrist-worn wearable	Multimodal algorithm	Diverse demographics	Ambulatory status	The algorithm accurately classifies ambulatory status across demographics
Pramanik et al.⁶⁰	2018	IoT and smart sensors	Not specified	Not applicable	Connected healthcare	IoT and smart sensors enable pervasive healthcare and remote monitoring

AIBSNF: Association of Information Brokers of Small and Medium-sized Businesses in the Netherlands framework; RTLSs: real-time location systems; ML: machine learning; AQ: aortic regurgitation; RA: rheumatoid arthritis; MB: multiband; AV: atrioventricular block; HRV: heart rate variability; ECG: electrocardiogram; PCA: principal component analysis; GPS: Global Positioning System; RMSE: root mean square error; OEPS: opto-electronic patch sensor; HR: heart rate; PPG: photoplethysmography; HMM: hidden Markov model; ICA: independent component analysis; CNN: convolutional neural network; RNN: recurrent neural network; LSTM: long short-term memory; GRU: gated recurrent unit; CSTS: cross-sectional time series; IoT: Internet of Things.

Activity recognition and motion monitoring

Human activity recognition (HAR) remains a principal application of smartwatch-ML systems. High classification accuracy was consistently observed across various studies, with decision tree classifiers achieving up to 99.63% accuracy in multi-sensor configurations.¹⁶ Random forest models, when coupled with principal component analysis (PCA), demonstrated 98.5% accuracy in classifying eight common daily activities.⁶¹ Hidden Markov models further enhanced HAR performance by leveraging dimensionality reduction via PCA and independent component analysis.⁴⁴

However, accuracy tended to diminish under real-world conditions where movement patterns were less structured and sensor placement varied. In clinical settings, smartwatches were successfully utilized to monitor inpatient mobility and bed exits with higher usability than conventional systems, although such deployments remained limited in scale and demographic diversity.³

Physiological monitoring and health metrics

Smartwatches have significantly advanced in capturing physiological signals. Commercial devices now offer heart rate and blood pressure measurements with clinically acceptable margins of error.¹⁴ HRV readings with under 1% measurement uncertainty, establishing their utility in stress and recovery assessments.⁴⁹

Emerging biosensors embedded in smartwatches expanded this capacity. An optical sweat sensor that accurately estimated sweat loss with a low root mean square error (RMSE), while real-time glucose monitoring via a self-powered wearable platform.^11,13 Electrochemical and textile-based biosensors further underscored the potential for non-invasive, continuous health surveillance, especially in geriatric and chronic care contexts.^10,29

ML integration and algorithm performance

ML integration was a central theme, with various architectures, decision trees, CNNs, RNNs, LSTMs, and GRUs employed to enhance data interpretation and decision-making. The AIBSNF integrates wearable biosensors with RTLSs to generate adaptive, personalized feedback loops.¹⁷

Notably, conventional ML models such as random forest and SVMs offered robust performance on structured datasets. However, deep learning models exhibited variable performance when applied to high-dimensional, noisy, or streaming physiological data.⁴⁵ Novel recommendation architectures like SASRec were more effective than CNN/RNN approaches in modeling sparse behavioral data for personalized fitness applications, signaling a shift towards context-aware personalization.⁴⁶

Clinical and exercise prescription relevance

Several studies bridged wearable technology with exercise science and clinical protocols. Benchmark criteria for cardiorespiratory fitness and stress testing, which served as references for ML-driven exercise prescription models.^20,21 The adaptability of ML allowed for dynamic calibration of exercise regimens based on real-time biofeedback, as opposed to static “one-size-fits-all” protocols.

Integration in rehabilitation settings also gained traction. Smartwatches provide reliable feedback on adherence and biomechanical recovery in orthopedic and cardiovascular rehabilitation.⁴⁷ However, scalability remained limited due to system complexity, data governance, and cost barriers.⁴⁰

User experience, adoption, and engagement

User engagement emerged as a pivotal determinant of long-term success. Eight motivational factors influencing smartwatch adoption, including perceived usefulness, enjoyment, and health consciousness.²⁷ The roles of app quality, relative advantage, and aesthetics in shaping user attitudes.²⁸

Nevertheless, several studies reported significant drop-offs in usage over time. Fitness behaviors follow non-linear, fractal patterns, suggesting declining engagement after initial adoption.⁵⁰ Visual feedback also influenced emotional response; simple monochrome graphs were associated with lower cognitive rumination than complex, multicolored visualizations, providing insights for designing ML-powered feedback systems.⁴⁸

Ethical, privacy, and technical constraints

Despite rapid technological advancement, systemic limitations persist. Ethical concerns surrounding data privacy, consent, and surveillance were prevalent but underexplored.^35,41 Users expressed a strong preference for granular data control and transparent consent mechanisms, which are often absent in current implementations.

Technical issues, particularly battery life, signal instability, and lack of interoperability, limit real-time performance and integration into healthcare ecosystems.^48,62 Standardization of wearable data streams remains an open challenge, impeding cross-platform applications and large-scale deployment.

GRADE assessment of selected studies

The GRADE evaluation revealed consistently high-quality evidence for most wearable technologies, especially in biosensing, activity recognition, and rehabilitation support. Moderate-quality evidence was observed in areas related to privacy and long-term user engagement. The findings are presented in Table 4.

Table 4.

GRADE assessment of evidence across studies on wearable technologies.

Study	Technology	Outcome	Effect	Quality of evidence	Recommendation strength
Balu and P, 2022¹⁶	Wearable multi-sensor system	Human activity recognition	99.63% accuracy using decision tree classifiers	High	Strong
Antognoli et al., 2023²²	Commercial smartwatches	Heart rate and blood pressure measurement	Clinical-grade accuracy demonstrated	Moderate	Moderate
Phatak et al., 2021¹⁷	AIBSNF	Data-driven healthcare and sports	Integration of wearable biosensors, RTLS, and ML	Moderate	Moderate
Auepanwiriyakul et al., 2020²³	Smartwatches	Inpatient motion monitoring	Accurate and user-friendly in hospital settings	Moderate	Moderate
L’Heureux et al., 2017²⁴	Big Data analytics	Machine learning challenges	Emerging techniques needed for effective analysis	Low	Weak
Najafabadi et al., 2015¹⁸	Deep learning	Big Data analytics	Struggles with streaming and high-dimensional data	Low	Weak
Gibson et al., 2019²⁰	Fitness assessment	Exercise prescription principles	Enhances program design and adherence	Moderate	Moderate
Fletcher et al., 2013²¹	Cardiovascular stress testing standards	Training and monitoring in clinical contexts	Established standards for guidance	High	Strong
Ross et al., 2019¹	Genetic and physiological variability	Individual exercise responses	Significant effects on exercise responses	Moderate	Moderate
Bayoumy et al., 2021²⁵	Smart wearable devices	Cardiovascular care	Integration and data standardization challenges	Moderate	Moderate
Kario, 2020²⁶	Wearable BP monitors	Anticipatory hypertension care	Enables stress-free, continuous monitoring	High	Strong
Dehghani, 2018²⁷	Netnography	Smartwatch usage intention	Reveals eight key motivational factors	Moderate	Moderate
Kim and Shin, 2015²⁸	AQ, RA, MB, and AV factors	Smartwatch adoption	Significantly affect perceived usefulness and ease of use	Moderate	Moderate
Majumder et al., 2017²⁹	Textile-based wearable sensors	Elderly health monitoring	Cost-effective tools for wellness tracking	Moderate	Moderate
Dias and Cunha, 2018³⁰	Vital sign monitoring systems	Wearable health devices	Growth in cardiovascular tech and ethical concerns	Moderate	Moderate
Sharma et al., 2021¹⁰	Electrochemical and optical biosensors	Continuous diagnostics	Driving wearable health monitoring forward	High	Strong
King and Sarrafzadeh, 2017³¹	Smartwatch feasibility studies	Remote health research	Lack of rigorous clinical validation	Moderate	Moderate
Henriksen et al., 2018³²	Fitbit, Garmin, and Apple	Consumer wearables	Dominated research usage from 2011 to 2017	High	Strong
Theurl et al., 2023³³	Smartwatches	HRV readings	Match ECGs for global HRV markers	Moderate	Moderate
Kristoffersson and Lindén, 2022⁷	Wearable sensor studies	Physical activity	Lack of demographic diversity and real-life conditions	Low	Weak
Balli et al., 2018⁶¹	Smartwatch sensor data	Daily activity classification	98.5% accuracy using PCA and random forest	High	Strong
Buke et al., 2015⁶³	Inertial sensor algorithms	Gait, fall, and sleep monitoring	Key algorithms highlighted for remote healthcare	Moderate	Moderate
Vorlíček et al., 2021⁶⁴	Garmin and Holux GPS	GPS accuracy	Garmin outperformed in city environments; Holux excelled in parks	Moderate	Moderate
Pavlov et al., 2023¹¹	Sweat loss estimation algorithm	Sweat loss prediction	High precision (RMSE 236 mL), outperforming existing models	High	Strong
Legner et al., 2019¹²	Wearable sweat sensors	Physiological and biochemical biomarkers	Non-invasive monitoring capabilities	High	Strong
Morresi et al., 2020⁶⁵	Smartwatch HRV readings	Measurement uncertainty	∼1% uncertainty, viable for physiological monitoring	High	Strong
Alzahrani et al., 2015⁶⁶	OEPS sensor	Heart rate monitoring	Accurate even during intense motion, surpassing commercial HR monitors	High	Strong
Salehizadeh et al., 2015⁶²	SpaMA algorithm	Motion-corrupted PPG signal reconstruction	Effective in improving HR monitoring during workouts	Moderate
Balu and P, 2022¹⁶	Wearable multi-sensor system	Human activity recognition	99.63% accuracy (decision tree)	High	Strong
Zhao et al., 2019¹³	Self-powered smartwatch with biosensor	Sweat glucose monitoring	Continuous, non-invasive glucose tracking	Moderate	Strong
Homayounfar et al., 2020⁶⁷	CNN on smartwatch usage	Battery discharge prediction	Accurate predictions from 832 users	High	Strong
Spann, 2016⁹	Policy commentary on wearables	Privacy protection	Urged legislation for stronger safeguards	Low	Conditional
King and Jessen, 2014³⁴	Smart meter privacy framework	Privacy risks	Identified major vulnerabilities	Moderate	Conditional
Anaya et al., 2017⁸	Wearable user privacy study	Consent and privacy expectations	Users demand clearer policies	Moderate	Strong
Datta et al., 2018³⁵	Literature review	Wearable privacy research	Urgent need for mitigation strategies	Moderate	Strong
Min et al., 2023³⁶	Sweat sensors	Health monitoring	Personalized and predictive potential	High	Strong
Guk et al., 2019³⁷	Wearables to implants	Diagnostics evolution	Real-time diagnostics potential	High	Strong
Soh et al., 2015³⁸	Review of wearable systems	Reliability and power issues	Key hardware/software limitations	High	Strong
Darwish and Hassanien, 2011³⁹	Sensor networks	Health monitoring	Good healthcare potential, design issues	Moderate	Strong
Patel et al., 2012⁴⁰	Wearables in rehab	Real-time feedback	Effective but lacks clinical integration	Moderate	Conditional
Vijayan et al., 2021⁴¹	Big data from wearables	Data management	Data overload and storage challenges	Moderate	Conditional
Ahmed et al., 2020⁴²	AI + EHR	Precision medicine	Enhanced clinical decision-making	High	Strong
Mannini and Sabatini, 2010⁴³	HMM + PCA/ICA + wearables	Activity recognition	High accuracy per individual	High	Strong
Zakeri et al., 2008⁴⁴	CSTS model + HR/accel data	Energy expenditure	Accurate youth energy prediction	Moderate	Strong
Kumar et al., 2023⁶⁸	k-Means clustering	Content recommendation	Outperformed other models	Moderate	Conditional
Kang and McAuley, 2018⁴⁵	SASRec model	Sequential recommendations	Outperformed CNN/RNN	High	Strong
Shiri et al., 2023⁴⁶	Deep learning models	Model comparison	CNN/LSTM/GRU varied by dataset; real-world limits	High	Conditional
De Fazio et al., 2023⁴⁷	Wearable rehab tools	Performance tracking	Useful but costly and clinician-dependent	Moderate	Conditional
Niess et al., 2020⁴⁸	Fitness tracker visuals	Goal reflection	Bar graphs reduce rumination	Moderate	Strong
Morris and Aguilera, 2012⁴⁹	Mobile/social/wearable tech	Mental health access	Expanded assessment and reach	High	Strong
Wang et al., 2022⁵⁰	Smart fitness wearables	Weight loss patterns	Effective but high drop-off rates	Moderate	Conditional

GRADE: Grading of Recommendations, Assessment, Development and Evaluation; AIBSNF: Association of Information Brokers of Small and Medium-sized Businesses in the Netherlands framework; RTLSs: real-time location systems; ML: machine learning; AQ: aortic regurgitation; RA: rheumatoid arthritis; MB: multiband; AV: atrioventricular block; HRV: heart rate variability; ECG: electrocardiogram; PCA: principal component analysis; GPS: Global Positioning System; RMSE: root mean square error; OEPS: opto-electronic patch sensor; HR: heart rate; PPG: photoplethysmography; HMM: hidden Markov model; ICA: independent component analysis; CNN: convolutional neural network; RNN: recurrent neural network; LSTM: long short-term memory; GRU: gated recurrent unit; CSTS: cross-sectional time series.

Discussion

The integration of smartwatches with ML algorithms presents a transformative potential for exercise prescription, physiological monitoring, and health behavior analysis. However, as this review reveals, much of the current literature, despite its technological sophistication, is constrained by methodological limitations, a lack of generalizability, and insufficient theoretical integration. This discussion aims to critically analyze these findings, identify unresolved research challenges, and articulate how this review contributes to advancing the field.

Activity recognition: Precision achieved, context ignored

The impressive classification accuracies (>98%) achieved in activity recognition are predominantly drawn from controlled, homogenous study conditions.^16,61 Such outcomes, while indicative of ML's raw analytical power, are rarely replicable in real-world settings characterized by demographic, environmental, and behavioral heterogeneity. This over-reliance on laboratory-derived datasets introduces a significant external validity issue. Very few studies included in this review systematically address algorithm robustness across varying body types, gait patterns, or movement artifacts induced by pathology (e.g. Parkinson's disease tremors).

Contribution

By synthesizing evidence across multiple sensor modalities and ML approaches, this study identifies a critical need for ecologically valid models trained on heterogeneous populations, a gap scarcely addressed in prior meta-analyses.

Physiological monitoring: Expanding capability, inconsistent fidelity

Smartwatches now exhibit the capability to track key physiological metrics such as heart rate, blood pressure, and HRV with clinically relevant accuracy.¹⁴ However, the fidelity of these measurements is not uniform across contexts. For instance, metrics such as HRV suffer from signal instability during movement or under conditions of excessive perspiration.³³ Furthermore, sensor degradation, placement variability, and interference from environmental conditions introduce substantial variability, which is rarely quantified or reported in the reviewed studies.

Research gap

There is a dearth of longitudinal validation studies evaluating sensor drift, calibration requirements, and reliability over time or across repeated physical stress exposures.

Contribution

This review uniquely compiles performance limitations specific to dynamic settings, ranging from intense exercise to clinical ambulation, emphasizing the need for standardized performance benchmarking frameworks for wearable physiological monitoring.

Ml integration: Efficacy without explainability

ML architectures such as random forests, CNNs, and LSTMs have demonstrated remarkable classification and regression capabilities. However, these models often function as “black boxes,” lacking transparency in decision-making. Few studies attempt to interpret feature importance, assess bias, or incorporate domain knowledge from exercise physiology into model design.

Research gap

There is insufficient application of explainable AI (XAI) techniques in the context of wearable fitness systems. This limits both clinician trust and user interpretability, two critical components of adoption in healthcare settings.

Contribution

This review advocates for a shift from performance-centric to interpretability-centric evaluations, especially for clinical use cases, and highlights emerging hybrid models (e.g. AIBSNF) that combine data-driven learning with rule-based contextualization.

Exercise prescription: Theoretical potential, fragmented implementation

Several studies propose smartwatch-guided fitness recommendations, but these lack consistency in how exercise parameters (e.g. frequency, intensity, time, and type (FITT) principles) are derived or adapted. Foundational works in exercise science are rarely integrated into ML pipelines. Moreover, rehabilitation applications remain largely theoretical, with limited studies demonstrating robust, outcome-based assessments in post-injury populations.²¹

Research gap

There is a pronounced disconnect between exercise science frameworks and ML-based exercise recommender systems. This undermines clinical and functional relevance.

Contribution

By mapping current technological capacities onto classical exercise prescription criteria, this review lays the groundwork for developing evidence-aligned, ML-personalized workout programs with real-time feedback loops.

User engagement and behavior: Measured data, missed context

Despite capturing vast physiological and behavioral data, smartwatches have limited success in sustaining user engagement. This is partially due to a lack of effective computing strategies, and models that consider emotional and motivational states in feedback design. Visualization choices and motivational congruence directly affect adherence, yet these insights remain underutilized.^48,50

Research gap

There is minimal integration of behavior change theory or adaptive user modeling in smartwatch feedback algorithms, resulting in high attrition rates and superficial personalization.

Contribution

This study brings together scattered evidence on affective feedback, motivation-driven design, and behavioral adaptation, calling for the incorporation of psychological personalization in future ML models.

Ethical and regulatory dimensions: A silent crisis

Ethical concerns surrounding data privacy, consent, and algorithmic transparency are underreported in the reviewed literature. As devices become more invasive and capable of real-time health inference, the lack of rigorous ethical oversight poses a growing risk.

Research gap

Few studies engage with GDPR-compliant frameworks, informed consent protocols, or secondary data use policies. There is almost no participatory co-design with target populations regarding data governance.

Contribution

By explicitly cataloging ethical omissions and contrasting them with best practices, this review underscores the urgent need for interdisciplinary collaboration to embed ethics into the design and deployment pipeline.^8,35 Table 5 shows a summary of key gaps and study contributions

Table 5.

Summary of key gaps and study contributions.

Gap identified	Contribution of this review
Over-reliance on lab-based validation	Emphasizes the need for ecologically valid datasets
Lack of sensor performance benchmarks	Calls for standardized reliability and calibration studies
Absence of explainable models	Advocates for the integration of XAI into health applications
Disconnection from exercise science theory	Proposes framework alignment with FITT principles
Superficial personalization	Highlights role of affective computing and behavior theory
Ethical and regulatory neglect	Establishes a roadmap for ethically aligned design

XAI: explainable artificial intelligence; FITT: frequency, intensity, time, and type.

This review provides a critical lens on smartwatch-assisted ML systems, identifying both technological potential and scientific oversight. The path forward demands a convergent research agenda, uniting computer science, biomedical engineering, behavioral psychology, ethics, and clinical expertise. Only through such interdisciplinary synthesis can wearable technologies achieve robust, responsible, and personalized health support.

Future directions

While smartwatch-assisted exercise prescription has shown considerable promise, several notable gaps remain in the current body of literature. First, there is a marked lack of standardized evaluation frameworks, which impedes meaningful comparison across studies. Many investigations employ disparate datasets, algorithms, and metrics, limiting the generalizability and reproducibility of findings. Furthermore, demographic representation is disproportionately skewed toward younger, tech-savvy populations from high-income countries, resulting in limited insight into usability and efficacy across age groups, socioeconomic strata, and individuals with disabilities or chronic conditions.

To advance the field, targeted recommendations can be made for key stakeholder groups:

For researchers, there is an urgent need to establish standardized data collection protocols and performance benchmarks for activity recognition, physiological monitoring, and ML integration. Additionally, longitudinal studies encompassing diverse populations are essential to understanding user adherence, algorithmic fairness, and health outcomes over time.

For developers, enhancing battery life, sensor precision, and real-time processing capabilities is critical for sustained and reliable use in everyday settings. Moreover, greater attention must be paid to user privacy and data protection, particularly through embedded privacy-by-design principles and on-device (edge) computing. The design of highly personalized and adaptive systems, which account for individual variability in physiological responses and preferences, should be prioritized.

For clinicians, the focus should shift toward the integration of smartwatch data into electronic health records (EHRs) and care protocols. Developing interoperability standards, validation procedures, and training resources will facilitate the clinical adoption of these technologies for monitoring, rehabilitation, and patient engagement.

For policymakers, robust data governance frameworks must be enacted to ensure the secure, ethical, and equitable use of wearable health data. Policies should address consent, data ownership, algorithmic transparency, and access rights, while also incentivizing the development and deployment of health-promoting digital tools in underserved populations.

As illustrated in Figure 2, each stakeholder group, researchers, developers, clinicians, and policymakers, has distinct yet interdependent priorities that must be addressed to realize the full potential of smartwatch-assisted exercise prescription. Addressing these gaps will require multidisciplinary collaboration across computer science, biomedical engineering, behavioral science, and public health domains. Without such coordinated efforts, the transformative potential of smartwatches in health and fitness may remain unrealized or inequitably distributed.

Figure 2.

Targeted future directions for key stakeholder groups in smartwatch-assisted exercise prescription. Recommendations focus on standardization, personalization, clinical integration, and ethical data governance. Multidisciplinary collaboration is essential to address current limitations and ensure equitable impact across populations.

Conclusion

This review has highlighted the rapidly evolving landscape of smartwatch-assisted exercise prescription, with a particular focus on the integration of ML algorithms for personalized workout recommendations and physiological monitoring. Empirical findings demonstrate high levels of accuracy in activity recognition and biomarker tracking, underpinned by increasingly sophisticated sensor technologies and analytical models. Moreover, applications across fitness, clinical, and rehabilitative contexts underscore the versatility and growing relevance of wearable systems in promoting health outcomes.

However, persistent challenges, ranging from algorithmic limitations and demographic biases to privacy concerns and fragmented validation standards, must be resolved to ensure equitable and effective implementation.

In summary, the convergence of smartwatches and ML represents a paradigm shift in personal health monitoring and fitness optimization. With careful attention to design, validation, and policy, this synergy has the potential to redefine preventive healthcare, empower individuals, and foster a more data-informed approach to wellness and rehabilitation.

Footnotes

ORCID iDs

Hassan Jubair

Mithela Mehenaz

Author contributions

Hassan Jubair: conceptualization, methodology, analysis, visualization, and writing and manuscript review. Mithela Mehenaz: literature search, reference formatting, and manuscript review.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

References

Ross

Goodpaster

Koch

, et al. Precision exercise medicine: understanding exercise response variability. Br. J. Sports Med. [Internet] 2019 Mar 12; 53: 1141–1153.

Smartphone and Smartwatch-based biometrics using activities of daily living. IEEE Journals & Magazine | IEEE Xplore. 2019. Available from: https://ieeexplore.ieee.org/abstract/document/8835065/.

Auepanwiriyakul

Waibel

Songa

, et al. Accuracy and acceptability of wearable motion tracking for inpatient monitoring using smartwatches. Sensors [Internet] 2020 Dec 19; 20: 7313.

Tamura

Chen

. Seamless healthcare monitoring [Internet]. Springer eBooks. 2017. Available from: https://doi.org/10.1007/978-3-319-69362-0.

Fidan

Huseynov

Ali

, et al. Recent inventions in additive manufacturing: holistic review. Inventions [Internet] 2023 Aug 11; 8: 103. Available from: https://www.mdpi.com/2411-5134/8/4/103.

The future of computing performance. Google Books. Available from: https://books.google.com/books?hl=en&lr=&id=83eskZJmRoMC&oi=fnd&pg=PA1&dq=Despite±these±impressive±advancements,±major ± gaps±remain.±While±machine±learning±offers±incredible±potential±for ± health±personalization,±traditional±algorithms±often±struggle±to±process±high-volume,±high-speed±data±such±as±real-time±sensor ± streams&ots=v0tNmvyCpW&sig=Bdm4_vflbPxmeV3_t9CppF3neF4

Kristoffersson

Lindén

. A systematic review of wearable sensors for monitoring physical activity. Sensors [Internet] 2022 Jan 12; 22: 73.

Anaya

LHS

Alsadoon

Costadopoulos

, et al. Ethical implications of user perceptions of wearable devices. Sci. Eng Ethics [Internet] 2017 Feb 2; 24: 1–28.

Spann

. Wearable fitness devices: personal health data privacy in Washington State. Seattle Univ. Law Rev [Internet] 2016 Aug 2; 39: 1411. Available from: https://digitalcommons.law.seattleu.edu/cgi/viewcontent.cgi?article = 2354&context = sulr.

10.

Sharma

Badea

Tiwari

, et al. Wearable biosensors: an alternative and practical approach in healthcare and disease monitoring. Molecules [Internet] 2021 Feb 1; 26: 48.

11.

Sweat loss estimation algorithm for smartwatches. IEEE Journals & Magazine | IEEE Xplore. 2023. Available from: https://ieeexplore.ieee.org/abstract/document/10061707/.

12.

Legner

Kalwa

Patel

, et al. Sweat sensing in the smart wearables era: towards integrative, multifunctional and body-compliant perspiration analysis. Sens. Actuators Phys [Internet] 2019b Jul 13; 296: 200–221.

13.

Zhao

Lin

, et al. A fully integrated and self-powered smartwatch for continuous sweat glucose monitoring. ACS Sens. [Internet] 2019 Jul 4; 4: 1925–1933.

14.

Metrological characterization of commercial smartwatches: Are these sensors suitable for the assessment of well-being? IEEE Conference Publication | IEEE Xplore. 2023. Available from: https://ieeexplore.ieee.org/document/10164071/

15.

Auepanwiriyakul

Waibel

Songa

, et al. Accuracy and acceptability of wearable motion tracking for inpatient monitoring using smartwatches. Sensors [Internet] 2020b Dec 19; 20: 7313.

16.

Balu

. Wearable multi-sensor data fusion approach for human activity recognition using machine learning algorithms. SSRN Electronic Journal [Internet]. 2022b Jan 1; Available from: https://doi.org/10.2139/ssrn.4014024.

17.

Phatak

Wieland

Vempala

, et al. Artificial intelligence based body sensor network framework—narrative review: proposing an end-to-end framework using wearable sensors, real-time location systems and artificial intelligence/machine learning algorithms for data collection, data mining and knowledge discovery in sports and healthcare. Sports Med. - Open [Internet] 2021 Oct 30; 7. https://doi.org/10.1186/s40798-021-00372-0

18.

Najafabadi

Villanustre

Khoshgoftaar

, et al. Deep learning applications and challenges in big data analytics. J. Big Data [Internet] 2015 Feb 23; 2. https://doi.org/10.1186/s40537-014-0007-7

19.

Machine learning with big data: challenges and approaches. IEEE Journals & Magazine | IEEE Xplore. 2017. Available from: https://ieeexplore.ieee.org/abstract/document/7906512/.

20.

Gibson

Wagner

Heyward

. Advanced fitness assessment and exercise prescription [Internet]. Human Kinetics eBooks. 2019. Available from: https://doi.org/10.5040/9781718220966.

21.

Fletcher

Ades

Kligfield

, et al. Exercise standards for testing and training. Circulation [Internet] 2013 Jul 23; 128: 873–934.

22.

Antognoli

Panni

Cosoli

, et al. Metrological characterization of commercial smartwatches: are these sensors suitable for the assessment of well-being? IEEE 2023: 6–10. DOI: 10.1109/metrolivenv56897.2023.10164071.

23.

Auepanwiriyakul

Waibel

Songa

, et al. Accuracy and acceptability of wearable motion tracking for inpatient monitoring using smartwatches. Sensors [Internet] 2020 Dec 19; 20: 7313.

24.

L’Heureux

Grolinger

Elyamany

, et al. Machine learning with big data: challenges and approaches. IEEE Access 2017; 5: 7776–7797.

25.

Bayoumy

Gaber

Elshafeey

, et al. Smart wearable devices in cardiovascular care: where we are and how to move forward. Nat Rev Cardiol [Internet] 2021 Mar 4; 18: 581–599.

26.

Kario

. Management of hypertension in the digital era. Hypertension [Internet] 2020 Aug 3; 76: 640–650.

27.

Dehghani

. Exploring the motivational factors on continuous usage intention of smartwatches among actual users. Behav Inf Technol [Internet] 2018 Jan 17; 37: 145–158.

28.

Kim

Shin

. An acceptance model for smart watches. Int Res [Internet] 2015 Jul 16; 25: 527–541.

29.

Majumder

Mondal

Deen

. Wearable sensors for remote health monitoring. Sensors [Internet] 2017 Jan 12; 17: 130.

30.

Dias

Cunha

JPS

. Wearable health devices—vital sign monitoring, systems and technologies. Sensors [Internet] 2018 Jul 25; 18: 2414.

31.

King

Sarrafzadeh

. A survey of smartwatches in remote health monitoring. J Healthc Inf Res [Internet] 2017 Dec 18; 2: 1–24.

32.

Henriksen

Mikalsen

Woldaregay

, et al. Using fitness trackers and smartwatches to measure physical activity in research: analysis of consumer wrist-worn wearables. J Med Int Res [Internet] 2018 Mar 22; 20: e110.

33.

Theurl

Schreinlechner

Sappler

, et al. Smartwatch-derived heart rate variability: a head-to-head comparison with the gold standard in cardiovascular disease. Eur Heart J Digit Health [Internet] 2023 Mar 23; 4: 155–164.

34.

King

Jessen

. Smart metering systems and data sharing: why getting a smart meter should also mean getting strong information privacy controls to manage data sharing. Int J Law Inf Technol [Internet] 2014 Mar 18; 22: 215–253.

35.

Datta

Namin

Chatterjee

. A survey of privacy concerns in wearable devices. In: 2018 IEEE International Conference on Big Data (Big Data), Seattle, WA, USA, 2018, pp. 4549–4553. DOI: 10.1109/BigData.2018.8622110.

36.

Min

, et al. Skin-interfaced wearable sweat sensors for precision medicine. Chem Rev [Internet] 2023 Mar 27; 123: 5049–5138.

37.

Guk

Han

Lim

, et al. Evolution of wearable devices with real-time disease monitoring for personalized healthcare. Nanomaterials [Internet] 2019 May 29; 9: 813.

38.

Soh

Vandenbosch

GAE

Mercuri

, et al. Wearable wireless health monitoring: current developments, challenges, and future trends. In: IEEE Microwave Magazine, May 2015 Vol. 16, no. 4, pp.55–70. DOI: 10.1109/MMM.2015.2394021.

39.

Darwish

Hassanien

. Wearable and implantable wireless sensor network solutions for healthcare monitoring. Sensors [Internet] 2011 May 26; 11: 5561–5595.

40.

Patel

Park

Bonato

, et al. A review of wearable sensors and systems with application in rehabilitation. J NeuroEng Rehabil [Internet] 2012 Apr 20; 9. DOI: https://doi.org/10.1186/1743-0003-9-21.

41.

Vijayan

Connolly

Condell

, et al. Review of wearable devices and data collection considerations for connected health. Sensors [Internet] 2021 Aug 19; 21: 5589.

42.

Ahmed

Mohamed

Zeeshan

, et al. Artificial intelligence with multi-functional machine learning platform development for better healthcare and precision medicine. Database [Internet] 2020 Jan 1; 2020. DOI: https://doi.org/10.1093/database/baaa010.

43.

Mannini

Sabatini

. Machine learning methods for classifying human physical activity from on-body accelerometers. Sensors [Internet] 2010 Feb 1; 10: 1154–1175.

44.

Zakeri

Adolph

Puyau

, et al. Application of cross-sectional time series modeling for the prediction of energy expenditure from heart rate and accelerometry. J Appl Phys [Internet] 2008 Apr 11; 104: 1665–1673.

45.

Kang

W-C

Mcauley

. Self-attentive sequential recommendation. In: 2018 IEEE International Conference on Data Mining (ICDM, Singapore, 2018, pp.197–206. DOI: 10.1109/ICDM.2018.00035.

46.

Shiri

Perumal

Mustapha

, et al. A comprehensive overview and comparative analysis on deep learning models. J Artif Intell [Internet] 2024 Jan 1; 6: 301–360.

47.

De Fazio

Mastronardi

De Vittorio

, et al. Wearable sensors and smart devices to monitor rehabilitation parameters and sports performance: an overview. Sensors [Internet] 2023 Feb 7; 23: 1856.

48.

Niess

Knaving

Kolb

, et al. Exploring fitness tracker visualisations to avoid rumination. Exploring fitness tracker visualisations to avoid rumination [Internet]. 2020 Oct 1. DOI: 10.1145/3379503.3405662.

49.

Morris

Aguilera

. Mobile, social, and wearable computing and the evolution of psychological practice. Prof Psychol ResPract 2012; 43: 622–626.

50.

Wang

Sohail

Tang

, et al. Impact of fractals emerging from the fitness activities on the retail of smart wearable devices. Fractals [Internet] 2021 Nov 11; 32. DOI: 10.1142/s0218348x2240112.

51.

Burns

Razmjou

Shaw

, et al. Adherence tracking with smart watches for shoulder physiotherapy in rotator cuff pathology: protocol for a longitudinal cohort study. JMIR Res Protocols [Internet] 2020 Apr 26; 9: e17841.

52.

Ren

, et al. Exploring cooperative fitness tracking to encourage physical activity among office workers. Proc ACM Human-Computer Interact [Internet] 2018 Nov 1; 2: 1–20.

53.

Ghanvatkar

Kankanhalli

Rajan

. User models for personalized physical activity interventions: scoping review. JMIR Mhealth and Uhealth [Internet] 2018 Oct 26; 7: e11098.

54.

Sullivan

Lachman

. Behavior change with fitness technology in sedentary adults: a review of the evidence for increasing physical activity. Front Pub Health [Internet] 2017 Jan 10; 4. DOI: 10.3389/fpubh.2016.00289.

55.

Keogh

Taraldsen

Caulfield

, et al. “It’s not about the capture, it’s about what we can learn”: a qualitative study of experts’ opinions and experiences regarding the use of wearable sensors to measure gait and physical activity. J NeuroEng Rehabil [Internet] 2021 May 11; 18. DOI: https://doi.org/10. 1186/s12984-021-00874-8.

56.

Banaee

Ahmed

Loutfi

. Data mining for wearable sensors in health monitoring systems: a review of recent trends and challenges. Sensors [Internet] 2013 Dec 17; 13: 17472–500.

57.

Netz

Yekutieli

Arnon

, et al. Personalized exercise programs based upon remote assessment of motor fitness: a pilot study among healthy people aged 65 years and older. Gerontology [Internet] 2021 Aug 17; 68: 465–479.

58.

Sartori

Llyod

Farina

. Neural data-driven musculoskeletal modeling for personalized neurorehabilitation technologies. In IEEE Transactions on Biomedical Engineering, vol. 63, no. 5, pp. 879–893, May 2016. DOI: 10.1109/TBME.2016.2538296. Available from: https://ieeexplore.ieee.org/abstract/document/7440844/

59.

Popham

Burq

Rainaldi

, et al. An algorithm to classify real-world ambulatory status from a wearable device using multimodal and demographically diverse data: validation study. JMIR Biomed Eng 2023; 8: e43726.

60.

Pramanik

PKD

Upadhyaya

Pal

, et al. Internet of things, smart sensors, and pervasive systems: enabling connected and pervasive healthcare. In: Elsevier eBooks [Internet]. 2018, pp. 1–58. DOI: 10.1016/b978-0-12-815368-0.00001-4.

61.

Ball

Sağbaş

Peker

. Human activity recognition from smart watch sensor data using a hybrid of principal component analysis and random forest algorithm. Meas Control [Internet] 2018 Nov 28; 52: 37–45.

62.

Salehizadeh SMA, Dao D, Bolkhovsky J, et al. A novel time-varying spectral filtering algorithm for reconstruction of motion artifact corrupted heart rate signals during intense physical activities using a wearable photoplethysmogram sensor. Sensors [Internet] 2015 Dec 23; 16: 10.

63.

Buke

Gaoli

Yongcai

, et al. Healthcare algorithms by wearable inertial sensors: a survey. In: China Communications, April 2015, vol. 12, no. 4, pp. 1–12. DOI: 10.1109/CC.2015.7114054.

64.

Vorlíček M, Stewart T, Schipperijn J, et al. Smart Watch versus classic receivers: static validity of three GPS devices in different types of built environments. Sensors 2021; 21: 7232.

65.

Morresi

Casaccia

Sorcinelli

, et al. Analysing performances of heart rate variability measurement through a smartwatch. In: 2020 IEEE International Symposium on Medical Measurements and Applications (MeMeA), Bari, Italy, 2020, pp. 1–6. DOI: 10.1109/MeMeA49120.2020.9137211.

66.

Alzahrani

, Hu S, Azorin-Peris V, et al. A multi-channel opto-electronic sensor to accurately monitor heart rate against motion artefact during exercise. Sensors 2015;15:25681–702. https://doi.org/10.3390/s151025681.

67.

Kumar

Pati

Deepa

, et al. Clustering the various categorical data: an exploration of algorithms and performance analysis. In: 2023 4th International Conference for Emerging Technology (INCET), Belgium, India, 2023, pp. 1–6. DOI: 10.1109/incet57972.2023.10170508

68.

Homayounfar

, Malekijoo A, Visuri A, et al. Understanding smartwatch battery utilization in the wild. Sensors 2020; 20: 3784.