Supervised machine learning-based ETX optimization for energy-efficient routing in IoT-enabled WSNs

Abstract

This paper addresses the challenge of energy-efficient and reliable data routing in Wireless Sensor Networks (WSNs) within Internet of Things (IoT) environments by optimizing the Expected Transmission Count (ETX) metric for efficient routing. Traditional ETX-based routing struggles with dynamic network conditions, leading to suboptimal path selection and increased energy consumption. To overcome these limitations, we propose a Machine Learning-Based ETX Optimization Approach, which dynamically adjusts ETX values based on real-time network conditions and historical transmission patterns. The approach employs a supervised learning model, specifically a CatBoost classifier, to predict the most energy-efficient and reliable routes. The model achieves a high classification accuracy of 98.9%, enabling precise differentiation between optimal and non-optimal links, thereby reducing retransmissions and balancing energy consumption across the network. Our approach is evaluated using extensive simulations, analyzing key performance metrics such as energy consumption, network lifespan, Packet Delivery Ratio (PDR), and communication overhead. Experimental results demonstrate that the proposed method significantly enhances routing efficiency, minimizes energy expenditure, and improves overall network performance. Specifically, our method improves network lifetime by 14.3%, energy efficiency by 16.7%, PDR by 26.4% and communication overhead by 8.06% compared to existing protocols. These results highlight the robustness and predictive power of our approach, making it a highly effective solution for integrating WSNs into IoT ecosystems while ensuring sustainable and efficient operation.