A lightweight deep evidence fusion framework for smart home appliance detection and classification via internet of...

In the Internet of Things (IoT) field, massive quantities of smart devices are connected, producing enormous data sizes that need developed management methods. One general task in a smart environment is the capability to precisely recognize and classify diverse kinds of objects using these methods. The evolution of technologies and advancements in structure brings several challenges, such as managing and controlling the entire process, ensuring protection at the server, security in smart homes, and more. A robust and flexible deep learning (DL) technique is presented to achieve appliance classification without restrictions on user behaviour. The approach reflects combined load reports throughout the shared process. DL presents great models for analyzing and processing IoT sensor data, permitting accurate appliance recognition. This study presents a Lightweight Deep Evidence Fusion Framework for Smart Home Appliance Detection and Classification (LDEFF-SHADC) model. The primary objective of the LDEFF-SHADC technique is to establish a robust framework for detecting and classifying smart home appliances using IoT device data. The LDEFF-SHADC approach performs data normalization using linear scaling normalization (LSN) to ensure uniformity and enhance processing efficiency. The improved snake optimization (ISO) model is utilized for the feature selection process to reduce dimensionality while preserving critical information effectively. In addition, a hybrid framework integrating gated recurrent units and the multi-head attention (GRU-MHA) classifier is implemented to classify and detect smart home appliances. The improved sparrow search optimization algorithm (ISSA) is employed to fine-tune the parameters of the GRU-MHA model and achieve enhanced optimization performance. The stimulated outcome investigation of the LDEFF-SHADC technique is examined under the IoT device recognition dataset. The performance validation of the LDEFF-SHADC technique portrayed a superior accuracy value of 98.90% over existing models.