Study Spotlight: Can AI Predict Whether Living Near a Cell Tower Will Give You Headaches?

Part of our Study Spotlight series — breaking down new EMF research into plain English. No jargon. No agenda. Just what the science says.

The Study at a Glance


📄 Title	A Decision Support System for Managing Health Symptoms of Living Near Mobile Phone Base Stations
📰 Journal	Journal of Biomedical Physics & Engineering (Feb 2026, Vol. 16, Issue 1)
🏫 Researchers	Parsaei H, Faraz M, Mortazavi SMJ — Shiraz University of Medical Sciences & University of Eastern Finland
🔗 DOI	10.31661/jbpe.v0i0.2310-1667
📊 PMID	41668986

Why This Is Interesting

This study sits at a fascinating intersection: EMF health concerns meet artificial intelligence. Instead of asking “does living near a cell tower make you sick?” (which decades of research hasn’t definitively answered), these researchers asked a different question:

“Can we build a tool that predicts WHO is likely to experience symptoms based on their living conditions?”

It’s a subtle but important shift — from causation to prediction.

Check your EMF exposure

See cell towers, power lines, and substations near any US address.

Search Your Address

What They Did

The Dataset

699 adults living near mobile phone base stations participated
Researchers collected 11 predictors related to participants’ living conditions (things like distance from tower, floor level, exposure duration, building materials, etc.)
They tracked 5 health symptoms: headache, sleep disturbance, dizziness, vertigo, and fatigue

The Machine Learning Approach

They tested two algorithms:

Support Vector Machine (SVM) — a classification algorithm that finds the optimal boundary between “symptomatic” and “asymptomatic” groups
Random Forest (RF) — an ensemble method that builds many decision trees and combines their predictions

They also compared against a previously published Multilayer Perceptron Neural Network (MLPNN) model for the same prediction task.

What They Found

The SVM model performed best. Here’s how accurately it predicted each symptom:

Symptom	Accuracy	AUC	Sensitivity
Headache	85.3%	0.99	70.0%
Sleep disturbance	82.0%	0.98	83.4%
Dizziness	84.0%	0.92	85.3%
Vertigo	82.4%	0.89	73.0%
Fatigue	65.1%	0.81	69.0%

For the non-data-scientists: an AUC of 0.99 is exceptionally high — it means the model could nearly perfectly distinguish between people who reported headaches and those who didn’t, based only on their living conditions. An AUC of 0.5 would be random chance.

The SVM model significantly outperformed the Random Forest and neural network approaches, particularly for fatigue (where the other models basically failed, with sensitivities of just 8-11%).

What This Means (and What It Doesn’t)

What This Means (and What It Doesn't)

What it means

Living conditions near cell towers are predictive of symptoms. Whether or not those symptoms are caused by RF exposure specifically, the model can identify patterns linking proximity/conditions to symptom reports.
The tool could be practically useful. Healthcare professionals or urban planners could use such a model to identify high-risk residential areas before people report problems.
Machine learning found patterns that traditional statistics might miss. The 11 predictor variables interact in complex ways — ML is well-suited for this kind of multi-variable analysis.

What it does NOT mean

This doesn’t prove cell towers cause these symptoms. The model predicts correlation, not causation. People living closer to towers might also experience more noise pollution, light pollution, stress about the tower, or other confounding factors.
Self-reported symptoms are inherently noisy. Participants knew they lived near cell towers. The nocebo effect — where awareness of a perceived hazard causes real symptoms — is well-documented in EMF research. People who believe cell towers are harmful are more likely to report symptoms.
699 participants is moderate, not large. And they were all from Iran, limiting generalizability to other populations and built environments.
The very high AUC values raise questions. An AUC of 0.99 is unusually high for predicting subjective health symptoms. This might reflect overfitting (the model memorizing the training data rather than learning generalizable patterns) or the predictors capturing more about health anxiety than about RF exposure effects.

The Bigger Picture: EMF and Machine Learning

This study is part of a growing trend of applying AI to environmental health questions. The approach has real potential:

Personalized risk assessment — instead of one-size-fits-all guidelines, ML models could factor in your specific building, floor, distance, and exposure duration
Urban planning tools — cities could use predictive models when deciding where to place cell towers
Research prioritization — ML can identify which variables matter most, directing future research

But it also has risks. A model trained on self-reported symptoms in a population that’s already concerned about cell towers will inevitably learn those concerns as features. Garbage in, garbage out — or more precisely, bias in, bias out.

Why EMF Radar Built Something Similar

Full disclosure: this study is directly relevant to what we do. EMF Radar calculates exposure scores for addresses based on tower proximity, frequency, and power — not entirely unlike what this ML model does, but using measured RF data rather than self-reported symptoms.

The key difference: we report measured and calculated exposure levels, not predicted health outcomes. We believe giving people objective data is more useful than predicting whether they’ll feel sick — because whether symptoms are caused by RF, by anxiety, or by something else entirely, everyone deserves to know what’s in their electromagnetic environment.

Strengths

Novel approach — first ML-based decision support system specifically for cell tower proximity symptoms
Multiple algorithms compared — SVM, Random Forest, and neural network benchmarked against each other
Practical orientation — designed as a tool for healthcare and policy, not just academic publication
Good sample size for a pilot ML study (699 participants)

Limitations

Self-reported symptoms with no blinding — nocebo effects uncontrolled
No actual RF measurements at participants’ homes — living conditions used as proxy
Cross-sectional design — single point in time, can’t establish causation
Single geographic region (Iran) — different building construction, cultural attitudes, and tower deployment patterns elsewhere
Potential overfitting — the very high AUC values warrant external validation
Conflict of interest note: One author (Mortazavi) is an editorial board member of the publishing journal, though the paper states he was not involved in peer review

What This Means for You

If you live near a cell tower and experience headaches, sleep issues, or dizziness, this study suggests those symptoms are statistically predictable — meaning you’re not alone, and your living conditions genuinely correlate with these experiences.

But “predictable” isn’t the same as “caused by cell tower radiation.” The honest answer is: we still don’t know whether RF exposure from cell towers at environmental levels directly causes these symptoms. What we do know is that your experience is real, the patterns are real, and the science is still working to understand why.

In the meantime, check your address on EMF Radar to see actual tower data for your location — because whatever the cause, knowing your exposure environment is the first step.

Study Details

Full citation: Parsaei H, Faraz M, Mortazavi SMJ. A Decision Support System for Managing Health Symptoms of Living Near Mobile Phone Base Stations. Journal of Biomedical Physics & Engineering. 2026 Feb;16(1):47-56. doi: 10.31661/jbpe.v0i0.2310-1667.

Open Access: Available via PubMed Central (PMC12883925)

Want to know what’s in your electromagnetic environment? Search your address on EMF Radar — no ML predictions needed, just real data.

Study: Can AI Predict Cell Tower Health Symptoms?