<![CDATA[Step detection is an essential feature in promoting healthy living through mobile applications. This study evaluates the accuracy of a Convolutional Neural Network (CNN) model implemented in the Physical Traces application for detecting steps based on accelerometer and gyroscope sensor data. Data were collected through experimental activities where 60 participants walked 20 steps and ran 10 meters, repeated three times each. The results show average accuracy exceeding 100%, indicating a tendency for overcounting. Evaluation was performed using Absolute Error, Relative Error, Symmetric Accuracy, and SMAPE. Statistical analysis (Mann-Whitney, Kruskal-Wallis), reliability test (Cronbach’s Alpha = 0.9178), and validity test (positive correlations) revealed significant differences by gender and age group. These findings indicate that CNN-based step detection works effectively, but improvements are necessary to address individual variability and real-world conditions.]]>
Copyrights © 2025
