An Attention-Based LSTM-ELECTRE Model for Intelligent and Proactive Load Balancing in Real-Time Fog Computing Environments

Abstract

Load balancing between nodes is a fundamental challenge for fog computing environments, which are associated with dynamic Internet of Things (IoT) applications that require real-time responses. From this perspective, this study proposes an advanced hybrid model that combines the deep learning capabilities of LSTM networks with an attention mechanism to focus on the most significant time segments of the data series when predicting the upcoming load. Then, a multi-criteria decision-making algorithm follows, the ELECTRE algorithm, allowing optimal load prediction and intelligent task distribution across nodes to achieve fog computing stability. The proposed model bridges the gap in previous studies that do not consider task distribution mechanisms across nodes, load balancing prediction, and the nonlinear nature of the data. The proposed model was tested on two Mendeley datasets and a synthetic dataset to evaluate its generalization performance. The proposed model demonstrated excellent performance in terms of accuracy (R2 up to 0.99), low prediction error (MAE = 0.0085), effective load balance, and 100% task completion rate. The results obtained, compared to other approaches, confirm that the proposed method demonstrates high accuracy, is generalizable, and can be integrated with other models, such as fault prediction systems. This makes it a promising framework for more stable and reliable fuzzy environments within the current timeframe.