EEG scalp data processing using FIR filter, Kaiser windowing technique with features extraction and classification for epileptic seizure detection

Epileptic seizures are caused by abnormal synchronized discharges of neuronal activity in the brain, which can be recorded from the scalp using multichannel EEG signals. Traditionally, neurologists interpret these signals manually, a process that is both time-consuming and dependent on their availability. This study proposes a fully automated method to distinguish seizure from non-seizure EEG signals using advanced signal processing and machine learning. The EEG signals are first filtered using a Kaiser window-based notch filter, which is particularly effective at reducing spectral leakage and removing line frequency noise (50–60 Hz). The performance of different Kaiser window β values (4, 6, 8, and 12) is analyzed, with β = 12 showing the best results due to minimal leakage, high side-lobe attenuation, and narrow main lobe width, indicating high filter selectivity. The filtered EEG signals are characterized using six statistical features: mean, mode, median, variance, kurtosis, and skewness. These features are then used to train a machine-learning classifier to detect epileptic activity automatically. The novelty of this work lies in the optimised filtering using the Kaiser window and its integration with statistical feature-based classification, providing an efficient, automated seizure detection system that minimises manual intervention.