Sensors Technologies for Human Activity Analysis Based on SVM Optimized by PSO Algorithm

The rapid growth of wearable sensors have increased the importance of human activity analysis in different areas of information technologies. Motion artifacts often degrade the performance of wearable sensors. Several wearable sensors have been used since the last decades in order to recognize physical activity detection. The wearable sensors could have numerous applications in medical and daily life routine activities like human gait analysis, health care, fitness, etc. In this paper, accelerometer and gyroscope sensors dataset has been used to propose an efficient model for physical activity detection. We designed a new feature extraction algorithm, Mel-frequency cepstral coefficient and statistical features to extract valuable features. Then, classification of different daily life activities is performed via Particle Swarm Optimization (PSO) together with SVM algorithm over bench mark motion-sense dataset. The results of our model shows that pre-classifier as PSO and SVM along with feature extraction module excel in term of accuracy and efficiency. Our experimental results have shown accuracy of 87.50% over motion-sense dataset. This model is recommended for the system associating in physical activity detection, especially in medical fitness field.