Treffer: Edge-Cloud Computing for IoT Data Analytics Using Optimized Progressive Graph Convolutional Networks.

The rapid proliferation of connected devices, including sensors, wearable and other Internet of Things (IoT) devices, has led to an explosion of time series data, driving advancements in human activity recognition (HAR). Deep learning (DL) based techniques have shown great promise in predicting human actions from such data, particularly from wearable sensors and mobile devices. However, traditional cloud-based processing of IoT data introduces latency and network congestion, while edge computing, although offering reduced latency, faces computational constraints for machine learning tasks. In this manuscript, Edge-cloud computing for IoT data analytics using optimized Progressive graph convolutional networks (ECC-IoT-PGCN-HSOA) is proposed. The objective of the proposed ECC-IoT-PGCN-HSOA approach is to enhance accuracy, reduce training time, and address the challenges associated with IoT data processing, such as latency and computational constraints. Initially, the input data are collected from the MHEALTH dataset. The input data is first pre-processed at the edge layer using Confidence Partitioning Sampling Filtering (CPSF) for data cleaning and normalization. This pre-processed data is then sent to the cloud, where it is fed into Progressive Graph Convolutional Networks (PGCN) to identify the type of activity in the HAR. The PGCN parameters are optimized with the Humboldt Squid Optimization Algorithm (HSOA) in the cloud layer, enhancing the accuracy of activity identification from sensor data. The proposed ECC-IoT-PGCN-HSOA approach is implemented in Python and the performance metrics like Accuracy, Precision, Recall, Matthews Correlation Coefficient (MCC), F1-score, and ROC are analysed. The proposed ECC-IoT-PGCN-HSOA approach achieves exceptional performance metrics, with an accuracy of 99.28%, a precision of 99.41%, and a RoC of 0.997. Furthermore, it significantly reduces the training time to 23.41 s, outperforming existing methods. These results demonstrate that the proposed ECC-IoT-PGCN-HSOA method is highly efficient, scalable, and capable of addressing the challenges associated with edge-cloud IoT data processing, ensuring accurate and timely human activity recognition. [ABSTRACT FROM AUTHOR]