Posture Prediction of Individuals Using Agricultural Machinery Under Whole-Body Vibration in a Lab Environment

Low back pain associated with exposure to whole-body vibration (WBV) is common among agricultural workers, and seated posture significantly affects health outcomes from WBV exposure. Current posture assessment methods rely on manual observation or body-worn sensors, which are labor-intensive and impractical for continuous monitoring. We developed a machine learning approach to classify seated posture using force sensors and accelerometers integrated into a vibration sensing seat pad for use in agricultural machinery, avoiding the need for body-worn sensors. Twenty-four participants were exposed to WBV in different upper body postures while seat pad force and acceleration data were recorded. We compared four machine learning architectures: Logistic Regression, Random Forest, Support Vector Machine, and Recurrent Neural Network with Gated Recurrent Unit (GRU). The GRU architecture substantially outperformed baseline models, achieving 89% accuracy (weighted F1 = 0.89) in classifying forward and backward leaning postures. To our knowledge, this study demonstrates the first application of machine learning to classify seated postures from seat pad force measurements during WBV exposure. Temporal modeling with an 18 s window proved essential for accurate classification, enabling non-invasive, continuous posture monitoring.