Seizures Prediction with Slower Alpha Rhythm and Faster Gamma Rhythm using CNN-XGBoost Fusion Algorithm

Electroencephalography (EEG) is a vital neurodiagnostic tool widely used for identifying brain-related disorders due to its non-invasive nature, high temporal resolution, and cost-effectiveness. However, manual analysis of EEG signals is often time-consuming, subjective, and prone to low inter-rater agreement, as it requires specialized training and experience. This study proposes an advanced deep learning framework that automates EEG signal analysis for emotion recognition and abnormality detection. The raw EEG signals are pre-processed using Short-Time Fourier Transform (STFT) to convert temporal data into RGB spectrogram images, highlighting key time–frequency patterns. These images are processed through a two-dimensional Convolutional Neural Network (2D-CNN) to extract high-level spatial and frequency-domain features. The extracted features are then classified using the Extreme Gradient Boosting (XGBoost) algorithm, which accurately predicts emotional states such as arousal, valence, and dominance. Additionally, the EEG signals are decomposed into core brainwave components—Alpha, Beta, Delta, and Gamma—using wavelet decomposition and Fourier techniques, which are further analyzed for detecting normal and abnormal neural activity. Support Vector Machines (SVM) are also employed as part of the classification pipeline to enhance model robustness. The proposed CNN–XGBoost fusion model demonstrates superior performance compared to traditional methods, offering a reliable and scalable solution for automated EEG interpretation in emotion analysis and neurological diagnostics.