Performance Evaluation of Machine Learning Models for Energy Optimization in Wireless Sensor Networks

Abstract

The proliferation of Wireless Sensor Networks (WSNs) as a backbone for the Internet of Things (IoT) has intensified the critical challenge of energy consumption, which directly governs network lifetime and operational efficacy. Traditional routing and optimization protocols often fall short in dynamically adapting to the complex, heterogeneous, and resource-constrained environments of modern WSNs. This paper presents a comprehensive performance evaluation of contemporary machine learning (ML) models applied specifically for energy optimization in WSNs. We systematically review and categorize ML-based approaches, including Reinforcement Learning (RL), Deep Learning (DL), and Bio-Inspired Swarm Intelligence, employed between 2021 and 2025 to address energy efficiency, cluster head selection, and routing path optimization. Utilizing a standardized simulation framework, we empirically evaluate prominent models—Deep Q-Network (DQN) for adaptive routing, a Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) hybrid for traffic and energy prediction, and a Modified Ant Colony Optimization (M-ACO) for pathfinding. Key performance indicators such as network lifetime, energy consumption per node, packet delivery ratio (PDR), and end-to-end delay are measured and compared against classical benchmarks like LEACH and HEED. The results demonstrate that ML models, particularly the DQN and M-ACO, achieve superior performance, extending network lifetime by 20-35% and improving energy efficiency by 15-25% compared to traditional protocols. The paper concludes by discussing the computational overhead trade-offs, scalability concerns, and future research trajectories for embedding lightweight ML solutions into WSN-IoT architectures for sustainable smart applications.