This paper provides a comprehensive analysis of barbell workout classification using gyroscope and accelerometer data from MetaMotion sensors. The project’s objective is to develop a trustworthy machine learning model that can precisely classify barbell exercises and count repetitions. The dataset consists of workouts performed by participants with various barbell exercises at the gym. Data pretreatment, feature engineering, predictive modelling, outlier detection, and repetition counting are all part of the study’s systematic methodology. Surprisingly, our method demonstrated the efficacy of our approach by yielding classification accuracies greater than 90% for a range of classification algorithms. The work enhances the disciplines of machine learning and quantified self by providing insights into accurate exercise classification using sensor data. The findings might have implications for customised fitness tracking and rehabilitation applications. This work lays the groundwork for future investigations into sensor-based activity detection and is also a valuable resource for academics and professionals involved in human activity recognition and machine learning.