Performance Analysis of Classifier for Human Daily Activities

Human action recognition system using wearable sensors for classification of human daily activities is presented. Data collected for six daily human activities using smart phone with the help of thirty subjects is used. Thirteen time domain and two frequency domain features are extracted from the sensor data. Two machine learning algorithms namely decision tree and support vector machine and one deep learning algorithm LSTM is used for classification. The performance of the three classifiers in classifying the daily activities is evaluated using the performance metrics precision, specificity, recall, f-score and accuracy. Simulation results shows that the overall accuracy of SVM is 99.30% which is 2.88% and 0.53% higher than decision tree and LSTM respectively.