Enhancing IoT Performance with an Explainable Deep Learning Algorithm for Latency-Critical Applications

A significant quantity of data has been generated as a consequence of the fast growth of IoT devices, which necessitates the use of effective processing in real-time Conventional cloud-based architecture are not ideal for speed-critical applications because they have many drawbacks, including high bandwidth expenditures, security risks, and latency concerns. The determination of this learning is to provide a comprehensible deep learning approach that can be implemented on fog nodes that are computing in order to improve Internet of Things efficiency while simultaneously lowering latency. The use of sensor-generated streams of information allows for the application of a LSTM model for the purpose of detecting anomalies and human behaviour in smart home surroundings. In order to guarantee speedier decision-making and reduce dependency on centralised cloud servers, the installation of LSTM on fog nodes is essential. Users are able to comprehend how decisions are made of the suggested model, which results in an increase in both confidence and disclosure. The model that was suggested has a comprehensibility mechanism. It is shown that the technique is 96.30% better in terms effectiveness in terms of decreased reaction time, decreased resource eating, and increased tolerance for faults. The methodology is verified by utilizing datasets that are accessible to the public and compared against centralised algorithms for deep learning. There is a 35% decrease in latency achieved by the fog-based LSTM model when compared to base on the cloud possibilities, according to the findings of the experiments. In the IoT, this work prepares the way for solutions that emphasize speed and trustworthiness, are scalable, and are driven by artificial intelligence (AI).