Abstract
This study presents a machine learning-based approach for automating workout tracking through wearable sensors. Using accelerometer data collected from a MetaMotion 10-axis sensor, various weightlifting exercises, such as bench presses, pull-ups, and squats, were identified and classified using MATLAB Classification Learner. Feature extraction focused on both time and frequency domain features, with an emphasis on optimizing classification performance by carefully selecting and adjusting the features used. Multiple machine learning algorithms were tested to determine the most effective classifier, with Support Vector Machines (SVM) proving to be the best-performing model, achieving an accuracy of 93.6%. The trained classifier was then implemented in a MATLAB application to predict exercise activities in real time, providing accurate time and set tracking for each workout. Future improvements could include integrating additional sensors, expanding the training dataset, and enhancing system robustness across multiple users.