Abstract
A wearable sensor system is worn at two different locations on the body to automatically classify different workout activities being performed by the trainees in the gym. The sensor provides raw acceleration data in the x, y, and z-axis, then imported into MATLAB. The classifier predicts the workout actions based on the time and frequency features extracted from the sensor data. The classifier used was a Quadratic kernel function for Support Vector Machine (SVM) using Bayesian optimization with 30 iterations. A training dataset with labels was used to train the SVM. The model was trained and tested using separate test data, and an average accuracy of 99% was obtained. Different sensor locations were compared and concluded that the wrist was the most preferred location for workout classification.