Rest and Effort Tremor Detection using Machine Learning
Tremor is an involuntary trembling movement or shaking that appears due to various factors such as a neurological disease, stress, some medicine, etc. It can also be caused by fatigue as a natural event after undergoing a long-time effort. In this paper, we focused on detecting various tasks effectuated by a group of participants wearing an accelerometer on a wrist to detect a tremor. During the experimental session, the participant performed a kind of movement. Two scenarios were deployed to compute the duration of each task. In the first case, duties were split on rest and effort. In the second scenario, the efforts were separated into two different activities.
We used various machine-learning techniques such as a sup- port vector machine (SVM) with different kernels, a tree decision (TD), ensemble classifier Bagging Tree (BT), and a k-nearest neighbours (k-NN) algorithm to classify and identify the rest and the action tasks. After detecting the classes corresponding to the events effectuated, we computed the time duration of effort during the study according to the windowing used. Statistical features were calculated from the three-axis accelerometer signal with different sliding window (128, 256, 320, and 512 window size) to train the algorithms. The best results were obtained by the Ensemble classifier. We achieved a 92.7% accuracy using a 320 length of the window for two classes and 92.2% for the second scenario.
Keywords - SVM, k-NN, Tree Decision, Ensemble Classifier, Activity Recognition, Rest and Effort