A systematic review of machine learning approaches for sleep posture recognition across sensing modalities...

Sleep posture directly affects sleep quality, respiratory function, and musculoskeletal health. Machine learning (ML) techniques have been increasingly applied to automatic sleep posture recognition across diverse sensing modalities. However, direct comparisons across existing studies remain difficult due to widespread reliance on small, private datasets and inconsistent evaluation protocols, which limit the generalizability of reported findings. This review aims to systematically characterize recent ML-based research on sleep posture recognition, identify methodological gaps, and assess deployment readiness across sensing modalities.

A systematic literature search was conducted across IEEE Xplore, Scopus, Web of Science, PubMed, and Google Scholar following PRISMA guidelines, covering studies published between January 2023 and December 2025. From an initial pool of 120 articles, 45 human-subject experimental studies met the inclusion criteria. Included studies were categorized into four sensing modalities: pressure-based, wearable sensor-based, vision- and imaging-based, and contactless radio-frequency (RF) and radar-based systems. Standardized data extraction covered dataset characteristics, preprocessing methods, learning models, sensing hardware, posture classes, and performance metrics.

Most studies focused on four primary posture classes - supine, prone, left lateral, and right lateral - and reported high accuracy, mostly under controlled laboratory conditions. However, subject-independent validation was not consistently applied, and few studies examined finer-grained posture taxonomies.

Recurring limitations, including private datasets and insufficient cross-subject validation, highlight the need for open benchmark datasets, standardized evaluation protocols, and deployment-oriented assessment to advance reliable and generalizable sleep posture recognition systems.