A hybrid DL with the Internet of Things to monitor human activities using wearable sensors

Abstract

A wide variety of physiological and functional information could be continually collected for applications in sports, wellness, and medicine due to the increasing popularity of wearable technology. Machine Learning (ML) is a potential technology for Big Data Analytics (BDA) used to rapidly classify and analyze this plethora of information. ML has proven successful in applications that use fast-growing software systems; however, resource limitations prevent ML from being fused on low-power portable devices. In this research, the ML method integrates characteristics through inertial detector information data set to enable effective or immediate segmentation. When node computing, the architecture of this hybrid system overcomes the disadvantages found in a traditional Deep Learning (DL) model. Before the information was sent to the Neural Network (NN) architecture, a pre-treatment of the spectral domain of the proposed approach for real-time computation in the Internet of Things (IoT) was performed. Using real-world laboratory and activity databases, our proposed DL strategy was compared with advanced techniques for improving classification accuracy. Performing better than other approaches our results test the capacity of the strategy across multiple data sets, including human activities and this method was consistent with the constraints of actual time analysis of mobile device nodes and a portable detection system.

Keywords

• Machine learning
• High-performance computing
• Human activity datasets
• Wearable sensor platform
• IoT