Poor Posture Detection Based on Behavioral and Environmental Data Using Machine Learning

Poor spinal posture is a common problem among primary school students, which may lead to long-term health issues. It is essential to detect poor posture and identify the key behavioral and environmental factors for its preventions and intervention strategies, such as the sitting time, the backpack weight, the physical activities, and the sleep habits. We develop a machine learning method using random forests to predict the incorrect posture occurrences based on the collected behavioral and environmental data of primary school students and identify important influential factors. The data collected from 782 primary school students include six categories of behavioral factors and five categories of environmental factors. Experimental results show that the prediction accuracy reaches 75.16%, and some of the most important factors for poor posture include carrying backpacks on two shoulders, the mattress firmness, backpack weights, and the excessive use of the electronic devices, which are able to facilitate the early detection and prevention of scoliosis for school children and provide guidelines and insights for families and schools to develop intervention strategies, such as using the right backpack, cutting down on the screen time, increasing physical activities, and making the sleep surface more supportive.