Energy-Aware and QoS-Enhanced Routing With Node Fault Prediction for Consumer IoT Networks Using ML Frameworks

The Internet of Things (IoT) has greatly influenced the development of consumer-focused smart environments, largely because of its ability to enable communication between devices. It enhances the overall experience of consumers in the IoT sector. Depending on the device mobility and operational conditions, IoT networks can be broadly classified as static or dynamic. In dynamic scenarios, where devices are mobile or operate in variable conditions, real-time data is crucial for decision-making by providing continuous insight into complex systems. Nonetheless, mobility introduces challenges like maintaining connectivity, optimizing routing, and ensuring timely decisions. The rapid growth of consumer IoT has further amplified the need for efficient routing and fault prediction to maintain reliability, as existing protocols often fail to jointly address the challenges of optimal routing and fault management comprehensively. Moreover, the integration of the Machine Learning (ML) model to handle these challenges is highly unexplored. To bridge these gaps, this paper proposes an energy-aware and QoS-enhanced data routing framework tailored for consumer IoT, integrating advanced fault prediction capabilities. We introduce a novel fault prediction method using the Auto-Encoder Anomaly Detection with Fuzzy (ADF) framework and an adaptive routing strategy powered by actor-critic reinforcement learning. Further, the results are compared with existing data routing techniques such as Dynamic Directional Routing (DDR), Enhanced Fuzzy based Energy Efficient Routing Protocol (E-FEERP), and Area Based Routing (ABR). Validated through real-field IoT dataset, our approach demonstrates improved performance and reliability, surpassing existing methods in medium- and large-scale networks. For instance, the proposed method achieves 82% data throughput over real-field datasets, outperforming existing methods. Moreover, it reduces average data latency to 1 seconds and enhances network lifetime up to 91%, while maintaining higher residual energy (88.5 J) compared to traditional methods under similar conditions.