Accurately detecting mental stress through electroencephalography (EEG) signals is challenging due to artifacts. Moreover, achieving optimal system performance in mental stress detection requires efficient processingand prominent feature extraction techniques. This study explores signal-denoising methods for stress detection using EEG signals, investigating various pre-processing and feature extraction techniques, including finite impulse response filters, wavelet transforms, and empirical mode decomposition. The comparative analysis underscores wavelet methodology as particularly effective for extracting stress-related features, achieving an 81.7% accuracy rate with support vector machines. Integration of optimized feature selection with genetic algorithms significantly improves model performance, yielding a notable 87% accuracy rate based on experimental findings. The study offers a practical framework for constructing efficient machine-learning models validated through MATLAB and Raspberry Pi implementations. It underscores their practical utility in real-world applications, especially in EEG-based stress detection methodologies.