Comparison of Several Classification Algorithms for Detecting Sitting Posture Based on MediaPipe Pose Extraction

Abstract

Poor sitting posture is a contributing factor to musculoskeletal and spinal health issues, especially for individuals who remain seated for extended periods. This study aims to compare the performance of several classification algorithms in detecting sitting posture based on image data extracted using MediaPipe Pose. The dataset consists of 483 sitting posture images, each labeled as either “good_form” or “bad_form.” Each image was processed to extract 132 pose features from 33 body landmarks. The data was then split into training, validation, and testing sets in a 60:20:20 ratio. Four classification algorithms were evaluated: Random Forest, K-Nearest Neighbor (KNN), Support Vector Machine (SVM), and Multi-Layer Perceptron (MLP). Experimental results show that the Random Forest algorithm achieved the highest accuracy at 82.47%. This study demonstrates that pose extraction using MediaPipe can be effectively utilized for automatic and accurate classification of sitting posture.