JMIR Publications

ABSTRACT

Background:

Accurate monitoring of moderate-to-vigorous physical activity (MVPA) is critical for advancing public health research and personalized interventions. Traditional accelerometery methods, reliant on regression-derived intensity cut-points, exhibit significant misclassification errors and poor generalizability to free-living environment. Recent advancements in machine learning (ML) and deep learning (DL) offer promising alternatives for automated MVPA detection.

Objective:

This scoping review synthesizes evidence on ML and DL techniques for MVPA estimation and/ or prediction using accelerometer data, focusing on performance, algorithm bias, sensor configurations, and translational potential.

Methods:

Following PRISMA-ScR guidelines, we conducted a systematic search across PubMed, IEEE Xplore, and Web of Science (1995–2025), supplemented by snowball sampling of selected reference lists.

Results:

Of 1,938 screened studies, 40 met inclusion criteria with 4 studies added by follow-up manual searches. While traditional ML models (e.g., Random Forest, Support Vector Magnitude) achieved strong laboratory performance with F1-score of 87.4-100% and accuracy of 87.9-100%. However, their real-world performance declined by 8-13.3% in F1-score and 6.6-12.2% in accuracy, due to environment noise and device heterogeneity. DL architectures (e.g., Convolutional Neural Networks, Transformers) achieved robust performance by leveraging raw signal dynamics with F1-score of 71.9-79.8% and accuracy of 87.9-100% in free-living settings. Hybrid models (e.g., CNN-BiLSTM) demonstrated state-of-the-art performance (F1-score: 91.4-98.4%, accuracy: 97.7-99.0%). Wrist-worn sensors dominated studies (75%) and matched hip/thigh placements in lab settings (mean F1-scores: 86.5–88.6%), but multi-sensor configurations (wrist + hip) yielded the highest accuracy (89.7%). Key challenges included algorithmic bias reducing applicability in elderly populations, and impaired reproducibility with only 42.5% of studies sharing code and data. Emerging opportunities are noted for edge computing and hybrid models integrating contextual data.

Conclusions:

ML and DL significantly enhance MVPA monitoring by automating feature extraction and improving adaptability to free-living variability. However, persistent gaps, such as limited generalizability, inconsistent validation protocols, and transparency deficits, hinder translation. Future research must prioritize inclusive model training, standardized reporting frameworks, and open science practices to realize the equitable potential of AI-driven physical activity assessment.