Integrating Intelligence at the Periphery with Deep Learning: Edge-Cloud Computing for IoT Data Analytics

Abstract

There is a deluge of data flowing across the network as a result of the proliferation of various linked devices, such as sensors, mobile, wearable, and other IoT devices. Machine learning (ML) operations often include moving data from the internet of things (IoT) to a central server, which increases network traffic and introduces delays. By bringing processing closer to the periphery of the network and the data sources, edge computing may solve such problems. However, ML tasks are not well-suited to edge computing because to its restricted processing capability. Thus, the purpose of this article is to explore ways to integrate cloud and edge computing in order to analyse data from the Internet of Things (IoT) by making use of edge nodes to minimise data transfer. Feature learning is executed on the adjacent edge node to process data close to the source, once sensors are grouped according to locations. We also take similarity-based processing into account while making comparisons. Machine learning is used to carry out feature extraction. The learned autoencoder's encoder component is stored on the edge, while the decoder component is stored in the cloud. Human activity recognition using sensor data was the task that was evaluated. The findings demonstrate that sliding windows, when used during the preparation phase, allow for data reduction on the edge of up to 80% without noticeably compromising accuracy