Lower back pain (LBP) is a prevalent health problem that is often effectively treated with physiotherapy/rehabilitation. Adherence to a prescribed rehabilitation program is positively correlated with its overall effectiveness. However, currently there are no simple and deployable quantitative methods to assess home-based LBP rehabilitation participation, where the majority of exercises should be performed. It is hypothesized that inertial data collected from a set of multi-IMU-based wearables analyzed with machine learning will successfully identify the performance of LBP exercises and good sitting posture. With an optimized system utilizing an XGBoost classifier and three IMU sensors placed at the lower back, thigh, and ankle, accuracies of 94% ± 3% and 90% ± 11% were achieved for classification of LBP exercises and good sitting posture, respectively. The technology generated within this project has the potential to improve the effectiveness of LBP rehabilitation by facilitating remote monitoring, early problem diagnosis, and quantitative clinical feedback.