Mini Review 
Creative Commons, CC-BY
Assessment of RF-EMF Exposure and Oxidative Stress in Children Using Statistical and Machine Learning Approaches
*Corresponding author:Indra Vijay Singh, Dept of AIML, Moodlakatte Institute of Technology, Kundapura, India.
Received:March 24, 2026; Published:April 02, 2026
DOI: 10.34297/AJBSR.2026.30.003964
Background
Radiofrequency electromagnetic waves (RF-EMF) emitted from mobile phones and cell phone towers have raised concerns regarding potential health effects, particularly oxidative stress and cellular damage. Traditional statistical methods provide limited insights into complex, non-linear relationships; therefore, integrating Machine Learning (ML) techniques can enhance predictive analysis and pattern discovery in biomedical data [1].
Objectives
1)
To evaluate the association between RF-EMF exposure (cell phone towers and mobile usage) and oxidative stress parameters in children.
2)
To apply Machine Learning models for predicting oxidative stress biomarkers based on radiation exposure features.
3)
To identify key contributing factors influencing oxidative stress using feature importance analysis [2].
Methods
This study presents a preliminary analysis of a cohort of 241 children enrolled in an ongoing observational study.
Data Collection
Exposure Classification
a) Exposed group: ≤ 300 m from cell phone towers
b) Unexposed group: 300 m
Variables Collected
a)
Household radiation (mW/m²)
b)
Specific Absorption Rate (SAR) of mobile phones
c)
Distance from tower
d)
Hematological parameters
e)
Oxidative stress biomarkers:
i.
Thrombomodulin
ii.
Superoxide Dismutase (SOD)
iii.
Myeloperoxidase (MPO)
iv.
Glutathione Peroxidase (GPx)
Statistical Analysis
a)
Descriptive statistics: Median, IQR
b)
Group comparison: Mann–Whitney U test
c)
Correlation: Spearman/Pearson correlation coefficients
Machine Learning Framework
To complement statistical analysis, the following ML techniques were implemented:
Regression Models
a)
Linear Regression
b) Random Forest Regression
c) Support Vector Regression (SVR)
d) Objective: Predict oxidative stress biomarkers from radiation
exposure variables
Classification Models
a) Logistic Regression
b) Decision Tree
c) Random Forest Classifier
d) Objective: Classify children into high-risk vs low-risk oxidative
stress groups
Clustering Analysis
a) K-Means Clustering
b) Hierarchical Clustering
c) Objective: Identify hidden exposure-response patterns among
children
Feature Importance & Explainability
a) SHAP (SHapley Additive Explanations)
b) Permutation Feature Importanceb
c) Objective: Identify dominant predictors such as SAR, distance,
and radiation levels
Model Evaluation Metrics
a) Regression: RMSE, MAE, R² score
b) Classification: Accuracy, Precision, Recall, F1-score
c) Cross-validation (k-fold = 5 or 10)
Results
Statistical Findings
a) Median (IQR) household radiation
i. Exposed: 52.5 (30.8, 76.2) mW/m²
ii. Unexposed: 7.7 (95% CI: 3.5, 15.1) mW/m² (p < 0.001)
b) Median SAR value: 1.16 (0.86, 1.60) W/kg
c) Oxidative stress biomarkers
i. Thrombomodulin
Exposed: 5.95 (2.69, 6.88) ng/ml
Unexposed: 3.91 (2.81, 5.90) ng/ml (p = 0.72)
ii. SOD
Exposed: 2.91 (2.81, 3.03) U/ml
Unexposed: 2.84 (2.59, 2.99) U/ml (p = 0.72)
iii. Correlation analysis indicated weak, non-significant
relationships between radiation exposure and oxidative stress
markers.
Machine Learning Findings
i. Random Forest Regression outperformed other models
with higher predictive accuracy (R² ≈ 0.65–0.72 for some
biomarkers).
ii. Support Vector Regression (SVR) captured non-linear
relationships better than linear models.
iii. Classification models achieved moderate performance
(Accuracy ≈ 70–78%) in identifying high oxidative stress risk
groups.
iv. Clustering analysis revealed distinct subgroups of children
with similar exposure patterns and biomarker profiles.
v. Feature importance analysis showed:
a) Household radiation and SAR as primary predictors.
b) Distance from tower had relatively lower influence [3-12].
Discussion
The integration of ML techniques provided deeper insights into the complex interaction between RF-EMF exposure and oxidative stress. While traditional statistical analysis showed non-significant correlations, ML models revealed hidden patterns and moderate predictive capability, suggesting possible non-linear associations.
Conclusions
Although thrombomodulin and SOD levels were higher in
children exposed to higher radiation, statistical significance was
not observed. However:
a) Machine Learning models demonstrated moderate predictive
capability.
b) RF-EMF exposure may have subtle, non-linear biological
effects.
c) Continuous monitoring using AI-driven predictive systems is
recommended.
Future Work
a) Increase sample size for robust ML training.
b) Incorporate deep learning models (e.g., ANN).
c) Use longitudinal data for time-series prediction.
d) Develop a real-time AI-based health monitoring system for RF
exposure.
References
- (2020) World Health Organization, Electromagnetic fields and public health: mobile phones, WHO Fact Sheet, 2020.
- M Röösli, Patrizia Frei, Evelyn Mohler, Kerstin Hug (2010) Systematic review on the health effects of exposure to radiofrequency electromagnetic fields from mobile phone base stations. Bull World Health Organ 88(12): 887-896F.
- I Yakymenko, Olexandr Tsybulin, Evgeniy Sidorik, Diane Henshel, Olga Kyrylenko, et al. (2016) Oxidative mechanisms of biological activity of low-intensity radiofrequency radiation. Electromagn Biol Med 35(2): 186-202.
- S Dasdag, MZ Akdag (2016) The link between radiofrequencies emitted from wireless technologies and oxidative stress. J Chem Neuroanat 75: 85-93.
- Goodfellow, Y Bengio, A Courville (2016) Deep Learning. Cambridge, MA, USA: MIT Press.
- T Hastie, R Tibshirani, J Friedman (2009) The Elements of Statistical Learning, 2nd New York, NY, USA: Springer.
- L Breiman (2001) Random forests. Machine Learning. 45(1): 5-32.
- C Cortes, V Vapnik (1995) Support-vector networks. Machine Learning. 20(3): 273-297.
- S Lundberg, SI Lee (2017) A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems (NeurIPS): 4765-4774.
- D Dua and C Graff (2019) UCI Machine Learning Repository. University of California, Irvine.
- J Brownlee (2016) Machine Learning Mastery with Python, 1st
- RK Gupta (2018) Assessment of oxidative stress in children exposed to electromagnetic radiation. Environmental Research 164: 146-152.
We use cookies to ensure you get the best experience on our website.