Proceedings of the Annual Meeting of the Cognitive Science Society

Individual alpha frequency (IAF) is a key marker of stable differences in neural processing speed and attention, and a critical parameter for frequency-tuned interventions such as transcranial alternating current stimulation (tACS), yet it is still often estimated with fragile, peak-picking heuristics. We propose a machine-learning algorithm that automates IAF detection in EEG data by (i) attenuating the 1/f background via exponential detrending, (ii) localizing candidate alpha peaks, and (iii) classifying them with a support vector machine using morphological features. On a longitudinal dataset (N = 204 young adults; frontal and parietal ROIs, two sessions), we benchmarked the method against consensus human ratings and the Philistine algorithm. The new approach halved localization error (MSE = 0.18 vs. 0.44) and reached 94% accuracy in detecting the presence of an IAF peak, while providing a continuous certainty measure. This enables expert-level, confidence-weighted IAF estimates for individualized cognitive and clinical protocols.