<![CDATA[Radiofrequency (RF) radiation from modern wireless devices has raised concerns about its potential effects on brain activity, especially in the Alpha and Beta frequency bands. However, research in this area faces significant challenges due to small and limited electroencephalogram (EEG) datasets, which often lead to inconsistent results and hinder reliable performance evaluation. This study addresses these limitations by incorporating synthetic data generation to enhance dataset robustness while maintaining realistic signal characteristics. The main objective of this study is to optimize the Power Asymmetry Ratio (PAR) for improved brainwave classification, using a dataset of 97 participants with engineering backgrounds from Universiti Teknologi MARA (UiTM), including both Males and Females, exposed to three types of RF exposure (Left Exposure (LE), Right Exposure (RE), and Sham Exposure (SE)) across two sessions (Before and During). The analysis incorporates advanced signal processing techniques, including ANOVA based feature analysis and K-Nearest Neighbors (KNN) modeling with Mahalanobis distance metric to evaluate gender-specific neural responses. Analysis of RF exposure data reveals clear gender-based patterns in classification performance. With a 70:30 data split for training and testing, in During exposure, Female participants achieved the highest KNN accuracy with 88% for training and 66% for testing when using Mahalanobis distance at K=3, while Male participants reached 80% (training) and 58% (testing) at K=2. The combined (synthetic + actual) data consistently outperformed actual data alone in training. This highlights synthetic data’s role in enhancing model robustness, especially for limited datasets. These results demonstrate that well-tuned classification algorithms can detect gender-specific brain responses to RF exposure, with Mahalanobis distance excelling in capturing subtle neural variations. The integration of synthetic data offers a scalable solution to small-data challenges, improving reliability in RF exposure studies while underscoring the need for gender- specific modeling in neural signals.]]>
