Supervised Machine Learning Algorithms for Bioelectromagnetics: Prediction Models and Feature Selection Techniques...

Area of Science:

• Computational biology
• Bioinformatics
• Electromagnetics

Background:

• High-performance computing and machine learning (ML) enhance predictive capabilities.
• Supervised ML offers efficient data pattern analysis for predictions.
• Assessing radiofrequency electromagnetic field (RF-EMF) biological impacts is crucial.

Purpose of the Study:

• To develop a novel supervised ML strategy for predicting RF-EMF effects on human and animal cells.
• To establish a predictive model without requiring in-vitro laboratory experiments.
• To analyze existing experimental data for patterns in cellular responses to RF-EMF.

Main Methods:

• Extracted data from 300 peer-reviewed publications (1990-2015) covering 1127 experimental case studies.
• Applied Principal Component Analysis (PCA) and Chi-squared feature selection to identify six optimal features.
• Utilized ten different classifiers and k-fold cross-validation for performance assessment, including accuracy, RMSE, AUC, and PRC Area.

Main Results:

• The Random Forest algorithm demonstrated superior performance across all metrics, achieving an Area Under the ROC Curve (AUC) of 0.903 with k-fold=60.
• Identified significant correlations between specific absorption rate (SAR) and frequency, and between SAR×time (cumulative exposure) and RF-EMF impact.
• Found no significant relationship between frequency and exposure duration.

Conclusions:

• Supervised ML, particularly Random Forest, provides a robust and accurate method for predicting RF-EMF effects on cells.
• SAR and cumulative exposure (SAR×time) are key determinants of RF-EMF biological impact.
• Future research with larger datasets can further refine predictive accuracy.