This study aims to reduce the dimensionality of traditional EEG features for Parkinson's disease (PD) diagnosis by proposing a channel selection method based on single-channel validation. The authors extracted 22 multi-scale features from the UNM and Iowa datasets using a 4th order Butterworth filter and wavelet packet transform. A total of 29 channels with the highest R2 scores were selected from 59 common channels. The proposed channel selection scheme achieved a 100% classification accuracy on the UNM dataset and demonstrated good generalizability on the Iowa dataset. The method significantly reduced the dimensionality of EEG feature vectors by 50%, and the highest classification accuracy was 100% for closed-eye state and 93.75% for open-eye state. The study highlights the effectiveness of the proposed channel selection approach in improving classification performance and reducing computational costs.This study aims to reduce the dimensionality of traditional EEG features for Parkinson's disease (PD) diagnosis by proposing a channel selection method based on single-channel validation. The authors extracted 22 multi-scale features from the UNM and Iowa datasets using a 4th order Butterworth filter and wavelet packet transform. A total of 29 channels with the highest R2 scores were selected from 59 common channels. The proposed channel selection scheme achieved a 100% classification accuracy on the UNM dataset and demonstrated good generalizability on the Iowa dataset. The method significantly reduced the dimensionality of EEG feature vectors by 50%, and the highest classification accuracy was 100% for closed-eye state and 93.75% for open-eye state. The study highlights the effectiveness of the proposed channel selection approach in improving classification performance and reducing computational costs.