Implementasi Sistem Deteksi Aktivitas Fisik Berbasis Android Menggunakan Temporal Convolutional Network dan Attention Mechanism

Abstract

The low level of physical activity in modern society has become a major risk factor for non-communicable diseases such as obesity, diabetes, and cardiovascular disease. This research aims to develop an Android-based physical activity detection system capable of automatically recognizing user activities in real-time. The system is designed to classify six categories of physical activity, walking, walking upstairs, walking downstairs, sitting, standing, and lying down, by utilizing data from the smartphone's built-in accelerometer and gyroscope sensors. The methodology employed in this study is a combination of the Temporal Convolutional Network (TCN) architecture and a Self-Attention Mechanism. Testing on the dataset shows that the model achieves an accuracy of 92.84% with a loss value of 0.3439. Furthermore, it obtained a precision of 93.36%, a recall of 92.94%, and an F1 Score of 92.86%, reflecting a consistent and reliable classification performance across all activity classes. Additionally, the confusion matrix indicates stable classification performance for all six activities. With its local inference capability, the system can operate without an internet connection and provide real-time feedback to the user. This research is expected to contribute to the development of practical, accurate, and efficient deep learning-based mobile solutions for physical activity monitoring to support a healthier lifestyle.