Human activity recognition and classification using of convolutional neural networks and recurrent neural networks

Abstract

Recently, by using the deep learning models, it has become easier to recognize the human activity with more accuracy than before by categorizing the activities that people are doing daily. Nowadays, with the extensive use of modern smartphones that have sensors, it has become easier to capture the data in raw format that has the movement details in three dimensions (X-Y-Z). In this paper, we utilized the open source WIreless Sensor Data Mining (WISDM) dataset which has six activities that are walking, jogging, standing, sitting, upstairs and downstairs. Each type of those activities consists of values in terms of (X, Y and Z) axes. We employed two types of deep learning algorithms that are Convolutional Neural Network (CNN) and Recurrent Neural Network - Long Short-Term Memory (RNN-LSTM). Our objective is to make a comparison between accuracy and loss after implementing the two models. We discovered that, when using the Convolutional Neural Network (CNN), the accuracy was 81%. However, the accuracy was 91% when using Recurrent Neural Network - Long Short-Term Memory (RNN-LSTM) and applying it on the same database. As a result, the Recurrent Neural Network - Long Short-Term Memory (RNN-LSTM) model outperformed the Convolutional Neural Network (CNN) model.