Sitting Posture Detection and Classification Using Machine Learning Algorithms on RapidMiner

Abstract

Integrating pressure sensors into cushion pads presents a viable posture monitoring and classification solution in innovative health care and ergonomic design. In this study, a cushion pad with a pressure sensor implanted that can recognize and classify different postures using machine learning techniques is developed and evaluated. The principal objective is to augment postural awareness and avoid disorders of the muscles. The cushion pad system was created and used by combining software algorithms with hardware sensors. Using a variety of machine learning approaches, RapidMiner, a data science platform, was used to analyze the pressure data to classify postures. The following algorithms are tested using cross-validation for a robust evaluation: Decision Tree, Naive Bayes, Neural Network, Random Forest, and K-Nearest Neighbors (K-NN). The outcomes showed that the various algorithms' levels of accuracy varied. The Naive Bayes algorithm demonstrated a lesser accuracy of 55.83% compared to the Decision Tree algorithm's 84.49% accuracy. The Random Forest algorithm surpassed the others with an accuracy of 85.98%, while the Neural Network approach produced an accuracy of 82.26%. The k-NN algorithm also yielded promising results, with an accuracy of 82.01%. According to these results, the Random Forest algorithm outperforms the Decision Tree algorithm for posture categorization in this specific example.  A workable approach for enhancing ergonomic health and avoiding posture-related illnesses is to integrate such machine learning models into a cushion pad with pressure sensor integration that can significantly help proactive posture management.