Machine Learning-Based Frequency-Selective RF-EMF Exposure Prediction in Mobile Communication Systems Using Spatial and Temporal Features

Abstract This study investigates machine learning models for the analysis and prediction of frequency-selective electric field exposure levels across multiple wireless communication services. The research is based on an extensive in situ measurement campaign conducted in the city of Prishtina and includes seven frequency bands corresponding to GSM, UMTS-LTE, and 5G wireless services. A diverse set of machine learning models, including linear regression, instance-based learning, neural networks, ensemble tree-based algorithms, and gradient boosting methods, was evaluated for electric field exposure prediction. To assess model robustness and geographic and temporal transferability, three validation strategies were considered: a conventional random hold-out framework, a buffered spatial framework designed to reduce the influence of geographically related observations, and a leave-one-date-out framework that withholds an entire calendar measurement date at a time to test generalization to previously unseen dates. The study presents a comprehensive comparison of predictive performance, computational efficiency, and feature importance across the investigated models and frequency bands. The results demonstrate that ensemble tree-based approaches achieve the highest predictive accuracy under random hold-out and buffered spatial validation, while temporal predictors emerge as the dominant factors influencing model performance. Furthermore, the analysis highlights the impact of validation strategy on predictive results and reveals differences in model robustness under geographically and temporally independent evaluation conditions. The findings demonstrate the potential of machine learning techniques for accurate frequency-selective RF-EMF exposure prediction and support their application in large-scale exposure assessment, monitoring, and wireless network planning activities.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-12
DOI
https://doi.org/10.21203/rs.3.rs-11202257/v1
Primary Topic
Electromagnetic Fields and Biological Effects
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Machine Learning-Based Frequency-Selective RF-EMF Exposure Prediction in Mobile Communication Systems Using Spatial and Temporal Features

Enver Hamiti, Hëna Maloku, Doruntinë Berisha, Mimoza Ibrani
Research Square
Electromagnetic Fields and Biological Effects
preprint

Machine Learning-Based Frequency-Selective RF-EMF Exposure Prediction in Mobile Communication Systems Using Spatial and Temporal Features

Enver Hamiti, Hëna Maloku, Doruntinë Berisha, Mimoza Ibrani
preprint en

Abstract

Abstract This study investigates machine learning models for the analysis and prediction of frequency-selective electric field exposure levels across multiple wireless communication services. The research is based on an extensive in situ measurement campaign conducted in the city of Prishtina and includes seven frequency bands corresponding to GSM, UMTS-LTE, and 5G wireless services. A diverse set of machine learning models, including linear regression, instance-based learning, neural networks, ensemble tree-based algorithms, and gradient boosting methods, was evaluated for electric field exposure prediction. To assess model robustness and geographic and temporal transferability, three validation strategies were considered: a conventional random hold-out framework, a buffered spatial framework designed to reduce the influence of geographically related observations, and a leave-one-date-out framework that withholds an entire calendar measurement date at a time to test generalization to previously unseen dates. The study presents a comprehensive comparison of predictive performance, computational efficiency, and feature importance across the investigated models and frequency bands. The results demonstrate that ensemble tree-based approaches achieve the highest predictive accuracy under random hold-out and buffered spatial validation, while temporal predictors emerge as the dominant factors influencing model performance. Furthermore, the analysis highlights the impact of validation strategy on predictive results and reveals differences in model robustness under geographically and temporally independent evaluation conditions. The findings demonstrate the potential of machine learning techniques for accurate frequency-selective RF-EMF exposure prediction and support their application in large-scale exposure assessment, monitoring, and wireless network planning activities.

Research Square
University of Prishtina (XK)
Industry, innovation and infrastructure, Good health and well-being
Electromagnetic Fields and Biological Effects
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon