Classification of human activity based on smartphone inertial sensor using support vector machine

The aim of this paper is to compare the performance of different kernel of classification support vector machine classification for classifying the physical daily living activities. Thirty subjects from a database performed activities such as walking, sitting, standing, laying, walking upstairs and downstairs. Inertial sensors signals ((accelerometer, gyroscope and magnetometer) from the smartphone are used to measure the human movements for each activity. The inertial sensor data were processed using signal processing method with several features of time domain and frequency domain. Multiclass support vector machine polynomial kernel (MC-SVM-Polynomial) and multiclass support vector machine Linear Kernel (MC-SVM-Linear) using One Versus All (OVA) methods were employed. These classification methods are assessed using the performance criteria such as precision percentage, recall percentage and correct accuracy classification rate percentage using 10-fold cross validation procedure. The results show that MC-SVM-Polynomial produces the best result with 98.57% compared to MC-SVM-Linear with 97.04% based on correct accuracy classification rate.