Predictive Modeling of Microstrip Antenna Parameters Through Feature-Enhanced Neural Network

Abstract

The rapid expansion of wireless communication networks has amplified the demand for efficient and accurate prediction of antenna performance metrics such as signal strength and power levels. Recent industry statistics show that over 90% of network service complaints in urban areas are linked to inconsistent signal quality, while global mobile data traffic is projected to grow by more than 25% annually over the next five years, intensifying the need for predictive performance optimization. Despite advancements, existing prediction models often rely on manual data analysis or limited-feature regression methods, which struggle with high-dimensional datasets and fail to capture complex non-linear relationships, resulting in suboptimal accuracy for real-world deployment. To address these challenges, this work introduces a machine learning–driven prediction framework for microstrip antenna performance that leverages preprocessing with correlated feature analysis to remove redundancy and enhance model efficiency. Existing approaches such as traditional linear regression and Lasso regression are incorporated as baseline models for benchmarking, ensuring a fair performance comparison. Building upon this, the proposed method implements a Multi-Layer Perceptron (MLP) regressor capable of capturing non-linear dependencies between input parameters and output performance metrics. The integration of feature correlation filtering prior to training ensures that the MLP focuses on the most informative attributes, significantly reducing noise and improving prediction stability. By targeting accurate estimation of both signal strength and power levels, this approach not only enhances the reliability of antenna performance forecasting but also provides a scalable solution adaptable to various deployment environments, supporting both industrial and research applications in next-generation wireless communication systems.