How Machine Learning Can Accurately Classify Postural Dysfunction through Activity Data

This study used machine learning models to predict whether middle-aged and older individuals had postural dysfunction based on a one-week period of physical activity data collected by a waist-worn accelerometer. The models achieved an area under the receiver operating characteristic curve (AUC) ranging from 0.67 to 0.73, with the support vector machine (SVM) and a gradient-boosted model, XGBoost, achieving the highest AUC of 0.73. Age was the most important variable for SVM classification, followed by accelerometer counts at various thresholds. The study suggests that ML analysis of accelerometer-derived physical activity data can be used to classify postural dysfunction in real-world environments such as the home.