Improving the accuracy of complex activities recognition using accelerometer-embedded mobile phone classifiers

Abstract

Using mobile phones for Human Activities Recognition (HAR) is very helpful in observing daily habit of the user and early detecting health diseases or accidents. Many studies have been published which have investigated the HAR with the help of mobile phones. However, these studies mainly focused on simple single locomotion activities. In real-world situations, human activities are often performed in complex manners. This study will investigate recognizing of complex activities with common classifiers those using in recognizing human activities. Data was collected about complex activities, then features were extract, finally the activities were classified. The experiment shows that the recognition accuracy of low level activities is higher than high level in all seven classifiers. Also it was noticed that the highest accuracy was got by IBK classifier (KNN). It got the highest accuracy in both positions, and in the three activity levels. Finally, the activities with armband position got more accuracy in all seven classifiers than in waist position. The study concluded that those classifiers are good to recognize low level activities (simple), but their performance reduces when the complexity of activities increase. So proper classifiers are needed to deal with the complex activities.