Continuous Monitoring and Modeling of High-Frequency EM Pollution Using Cutting-Edge Regression Techniques

Electromagnetic (EM) sources have been of scientific interest for decades, primarily because of their intricate wave structures and the ways they travel through various media. Both national and international agencies regulate EM emissions across different frequency bands, making sure the public remains safe, mostly by setting limits on electric field intensity and power density. Here, we propose a precise predictive model for ongoing, high-frequency EM pollution, using data collected over six months in a mid-sized metropolitan area. Measurements were taken at a rate of 2 Hz, 24 hours a day. To forecast electric field intensity, we tested five regression models: Liquid Time-Constant Networks (LTC), Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) Networks, Gated Recurrent Unit (GRU) Networks, and Kolmogorov-Arnold Networks (KAN). Among these, LTC achieved the top test accuracy of 97.95%, although the other models also proved highly reliable.