Machine Learning Methods in Posture-Related Applications in Children up to 12 Years Old: A Systematic Review

One of the most important factors in how infants and young children learn to move is postural control. This systematic review aims to evaluate the machine learning methods in posture-related applications for children aged 0-12. Following PRISMA guidelines, we systematically searched the PubMed, Web of Sciences, SCOPUS, and ProQuest Central databases. Twenty-two studies were included in the qualitative synthesis following screening of 199 articles, with methodological quality assessed as moderate to good using the MINORS scale (scores ranging from 8/16 to 19/24). The reviewed research involved diverse samples of infants and children up to 12 years old, employing sensor-based technologies such as inertial measurement units, force plates, pressure mats, and video cameras to extract kinematic and postural features for machine learning applications. Reported accuracies, typically exceeding 85%, reflected considerable methodological heterogeneity related to sensor modality, data quality, and model architecture. Algorithms such as Random Forest, SVM, and CNN were most frequently and effectively applied for posture classification, early detection of developmental delays, and diagnosis of conditions such as cerebral palsy and autism spectrum disorder, demonstrating promising potential for at-home monitoring and clinical interventions.