Discriminative EEG Feature Extraction Using the Adaptive Synchrosqueezing Wavelet Transform for Epileptic Seizure Detection

Abstract

This study investigated using the adaptive synchrosqueezing wavelet transform (ASST) for feature extraction in EEG-based epileptic seizure detection. Traditional time-frequency methods, notably the continuous wavelet transform (CWT) and short-time fourier transform (STFT), were used to extract similar features and compared against ASST for seizure detection performance. Classification experiments using support vector machines (SVM), random forest (RF), and K-nearest neighbors (KNN) showed that ASST-derived features led to strong classification performance for distinguishing seizure and non-seizure EEG segments. The random forest classifier achieved the best performance using ASST features, with an accuracy of 99.09%, an F1-score of 97.71%, and an AUC-ROC of 0.9990 after applying class balancing with SMOTE. The results showed that ASST achieved slightly higher classification performance compared to STFT and CWT, suggesting its potential as an effective and adaptive tool for non-stationary EEG analysis in seizure detection.