Sit-to-Stand Phases Detection by Inertial Sensors

The Sit-to-Stand(STS) is defined as the transition from the sitting to standing position. It is commonly adopted in clinical practice because musculoskeletal or neurological degenerative disorders, as well as the natural process of ageing, deter-mine an increased difficulty in rising up from a seated position. This study aimed to detect the Sit To Stand phases using data from inertial sensors. Due to the high variability of this movement, and, consequently the difficulty to define events by thresholds, we used the machine learning. We collected data from 27 participants (13 females,24.37±3.32 years old). They wore 10 Inertial Sensors placed on: trunk,back(L4-L5),left and right thigh, tibia, and ankles. The par-ticipants were asked to stand from an height adjustable chair for 10 times. The STS exercises were recorded separately. The starting and ending points of each phase were identified by key events. The pre-processing included phases splitting in epochs. The features extracted were: mean, standard deviation, RMS, Max and min, COV and first derivative. The features were on the epochs for each sensor. To identify the most fitting classifier, two classifier algorithms,K-nearest Neighbours( KNN) and Support Vector Machine (SVM) were trained. From the data recorded, four dataset were created varying the epochs duration, the number of sensors. The validation model used to train the classifier. As validation model, we compared the results of classifiers trained using Kfold and Leave One Subject out (LOSO) models. The classifier performances were evaluated by confusion matrices and the F1 scores. The classifiers trained using LOSO technique as validation model showed higher values of predictive accuracy than the ones trained using Kfold. The predictive accuracy of KNN and SVM were reported below: • KFold – mean of overall predictive accuracy KNN: 0.75; F1 score: REST 0.86, TRUNK LEANING 0.35,STANDING 0.60,BALANCE 0.54, SITTING 0.55 – mean of overall predictive accuracy SVM: 0.75; F1 score: REST 0.89, TRUNK LEANING 0.48,STANDING 0.48,BALANCE 0.59, SITTING 0.62 • LOSO – mean of overall predictive accuracy KNN: 0.93; F1 score: REST 0.96, TRUNK LEANING 0.79,STANDING 0.89,BALANCE 0.95, SITTING 0.88 – mean of overall predictive accuracy SVM: 0.95; F1 score phases: REST 0.98, TRUNK LEANING 0.86,STANDING 0.91,BALANCE 0.98, SIT-TING 0.92