KLASIFIKASI POSTUR DUDUK BERBASIS CONVOLUTIONAL NEURAL NETWORK (CNN) UNTUK EVALUASI RESIKO ERGONOMI

AHMAD SANUSI (Universitas Faletehan)
Dede Brahma Arianto (Universitas Faletehan)

Jurnal Informatika dan Teknik Elektro Terapan
Vol. 14 No. 3 (2026)

Improper sitting posture is one of the leading causes of musculoskeletal disorders among workers and students. Automatic sitting posture detection using artificial intelligence has the potential to serve as an effective and efficient monitoring solution. This study develops a sitting posture classification system based on deep learning using the MobileNetV2 architecture with a transfer learning approach. The dataset consists of 938 images across three posture classes, namely good_posture, forward_lean, and backward_lean, obtained from the Roboflow Universe platform. Training was conducted in two phases using a progressive fine-tuning strategy with optimization mechanisms including EarlyStopping, ModelCheckpoint, and ReduceLROnPlateau. Experimental results show that the model achieved a best validation accuracy of 97.33% and a test accuracy of 95.74% with a macro F1-score of 0.9565. The resulting model is lightweight and has been converted to TFLite format, making it ready for deployment on mobile devices. This study demonstrates that MobileNetV2-based transfer learning can accurately classify sitting postures even with a limited dataset, and has strong potential for further development as a real-time ergonomic monitoring system.