Deep CNN-GRU Based Human Activity Recognition with Automatic Feature Extraction Using Smartphone and Wearable Sensors

Abstract

This article describes a method to Human Activity Recognition (HAR) challenges based on data from wearable and smartphone sensors. We introduced a deep learning model and recognition system that is a combination of CNN (Convolutional Neural Network) and GRU (Gated Recurrent Unit) to improve results. Preferably, the data have been collected from several wearables as the participants go about their everyday activities. The convolutional neural network (CNN) deployed to improve the extraction of features at various scales. The derived attributes are then inserted into the gated recurrent unit (GRU), which labels features and enhances feature representation by understanding temporal connections. The CNN-GRU model uses a fully inte-grated (FC) layer, which is employed to hook up the feature maps with the classification standard. Three publicly accessible datasets, UCIHAR, OPPORTUNITY, and MHEALTH, were used to test the model's performance, with accuracy rates of 98.74%, 99.05%, and 99.53%, respectively. The outcomes show that the proposed model transcends some of the notified results in terms of activity detection.