| مستخلص: |
The exploration of human physical activities has long been a subject of scientific inquiry. Researchers have consistently delved into the study of human activities, motivated by their myriad benefits across domains like physical health and sports. The advancements in electronic systems, coupled with the advent of diverse sensors, both portable and wearable, have resulted in the generation of substantial datasets like WiSDM, HAR, ...etc. This influx of data has played a pivotal role in the development of artificial intelligence-based systems for the recognition and classification of these activities. At this stage, this study concentrated on classifying six human physical activities sourced from the WiSDM dataset. Four distinct methods (Time Domain, Spectral Domain, Fractal dimension, and Haar Wavelet Transform) were employed to extract features from individual accelerometers after segmenting the signal of each activity through a sliding window. Subsequently, the features with the most significant impact were selected using gray wolf optimization for training the Multi-Class Support Vector Machines (MC-SVMs). The test outcomes revealed that the incorporation of Gray Wolf optimization with Multi-Class Support Vector Machines (MC-SVMs) enhanced the classification accuracy for the examined activities. The suggested system had shown an exceptionally good performance, where it had achieved precision, accuracy, F1-Score and recall rates of 99.45%, 99.44%, 99.45%, and 99.46% respectively. [ABSTRACT FROM AUTHOR] |