Imeh Umoren, Saviour Joshua Inyang, Ubong E. Etuk, Daniel Essien

Abstract

Electromagnetic Field (EMF) radio frequency exposure is a growing concern due to its impacts on public health and the environment. This study aims to develop a data-driven framework for clustering and analyzing long-term far-field EMF exposure in Uyo Metropolis, Nigeria, with a focus on identifying exposure patterns and assessing their implications. Data were measured at multiple locations using smart meter strategically deployed across three major roads in uyo metropolis to capture variations in exposure levels. The preprocessing steps involved data cleaning and normalization to enhance data quality and reliability for meaningful analysis.  Four clustering algorithms, namely, K-Means, Hierarchical Clustering, DBSCAN, and Gaussian Mixture Model (GMM), were employed to analyze the distribution of radiation levels. The Silhouette score was used to evaluate the different clustering methods with respect to cohesion within clusters and separation from other clusters. The best results were obtained by Hierarchical Clustering and GMM, each achieving a mean Silhouette score of 0.81, indicating well-defined and highly contrasting clusters. K-Means performed moderately well, with an average Silhouette score of 0.73, while DBSCAN, due to its sensitivity to noise and parameter settings, achieved a lower score of 0.62. These findings highlight significant spatial variability in EMF exposure across different urban zones, emphasizing the need for targeted regulatory measures. The study underscores effectiveness of machine learning and offers a scalable approach for characterizing EMF exposure. Results reported offer scalable and data-driven framework for characterizing exposure patterns, with important implications for public health policies, urban planning strategies, and regulatory interventions.