Using machine learning methods to detect physical conditions with postural balance

Previous researches investigated the association between diseases and the postural balance (PB), such as Parkinson's disease, multiple sclerosis, and Leprosy, etc. However, there is limited study exploring whether the PB can predict a person’s physical condition. Therefore, the aim of this study was to build a physical conditions detection system via a simple machine learning classifier–logistic regression (LR) with PB characterized by the center of pressure (COP) measured by a force plate. We converted COP to total excursion distance (TOTEX), TOTEX of anterior–posterior distance (TOEXAP) and TOTEX of medial–lateral distance (TOTEXML) as major features in the LR model along with gender, age, and body mass index (BMI). We conducted a perspective study to collect 67 patients’ records. Using those 67 records, we built 6 independent LG models based on gender, age, BMI, and collaborated with and without PB measurements to examine the effectiveness of using PB in the model to predict a person’s physical condition. We compared those 6 LR models’ performances based on the Area Under the Receiver Operating Characteristics (AUC), confusion matrix including accuracy, sensitivity, and specificity rate. The performance comparison results showed the predictive models with PB measurements were better than those of without PB (average AUC: 0.81 vs. 0.72). Therefore, the proposed physical conditions detection system can better discriminate healthy and unhealthy person with PB measurements in the LR classifier.