Human activity recognition by smartphone

Human activity recognition is one of the most important core building blocks behind many applications on smartphone such as medical applications, fitness tracking, context-aware mobile, human survey system, etc. This paper describes a robust system for human activity recognition by smartphone. Different from other work, we investigated the use and combination feature selection and instance selection to reduce dimensionality of dataset in order to enhance the performance. We implemented the system on Android and our experimental results showed that our system achieves better accuracy of up to 15% and the response time is 3 to 5 times faster when comparing to the original system.