Window Selection Impact in Human Activity Recognition
Nurul Retno Nurwulan, Bernard C. Jiang · DOAJ (DOAJ: Directory of Open Access Journals) · 2020
Signal segmentation is usually applied in the pre-processing step to make the data analysis easier. The signal data is divided into small epochs with same characteristics such as frequency. Windowing approach is commonly used for signal segmentation. However, it is unclear which type of window should be used to get optimum accuracy in human activity recognition. In this study, the recognition accuracy of each window types is evaluated and compared to determine the impact of window selection in human movement data. From the evaluation, the overlapping 75% window with 0.1 s length provides the highest accuracy with mean, standard deviation, maximum, minimum, and energy as the features. Shannon entropy works better as a feature when J48 is the classifier.