Human Activity Recognition Using Smartphone Sensor Based on Selective Classifiers
Mst. Alema Khatun, Mohammad Abu Yousuf
Abstract
HAR, elaborated as Human Activity Recognition, is a motile procedure of recognizing different physical activities performed by the human being in different environments. Walking, jogging, running, stair-up, stair-down, etc. are some of the examples of such kinds of actions. Our primary goal is to reckon different actions performed by a human subject. Although several techniques exist to identify different activities, we use Smartphone accelerometers because of the availability and ease of use and at the same time state-of-the-art technology. Here, we are approaches to using machine learning and deep learning techniques on the publicly available datasets. Random forest, Support vector machine, and CNN have been used to analyze and compare the performance. Although there is a lot of work to be done so far, we want to show that deep learning has better results than machine learning. A comparative study is done on these three classifiers using different accuracy measurements like accuracy rate, performance, and so on. Compared to all these three classifiers, experiment results show the CNN model can identify human activity with a good accuracy rate of 99% and with high performance.