This study proposes a multi-scale feature and multi-channel selection approach for Parkinson's disease (PD) diagnosis using electroencephalogram (EEG) signals. The objective is to reduce the dimensionality of traditional EEG features by selecting the most informative channels based on single-channel validation. The method involves extracting 22 multi-scale features from the UNM and Iowa datasets using a 4th order Butterworth filter and wavelet packet transform. Based on single-channel validation, 29 channels with the highest R2 scores were selected from a pool of 59 common channels. The proposed channel selection scheme was validated on the UNM dataset and tested on the Iowa dataset to assess its generalizability. The results show that the proposed model achieves an optimal classification accuracy of 100% on the UNM dataset and 93.75% on the Iowa dataset. The channel selection method significantly reduces the dimensionality of EEG feature vectors related to PD by 50%. The study also demonstrates the effectiveness of the proposed method in both the UNM and Iowa datasets, with the highest classification accuracy of 100% for the closed-eye state and 93.75% for the open-eye state. The results indicate that the proposed channel selection approach is effective in improving classification performance and generalization capability. The study also highlights the importance of channel selection in reducing the complexity of EEG data for PD diagnosis. The findings suggest that the proposed method can be applied to other EEG-related tasks, such as epilepsy diagnosis and emotion recognition, to enhance its applicability and utility in various clinical and research settings.This study proposes a multi-scale feature and multi-channel selection approach for Parkinson's disease (PD) diagnosis using electroencephalogram (EEG) signals. The objective is to reduce the dimensionality of traditional EEG features by selecting the most informative channels based on single-channel validation. The method involves extracting 22 multi-scale features from the UNM and Iowa datasets using a 4th order Butterworth filter and wavelet packet transform. Based on single-channel validation, 29 channels with the highest R2 scores were selected from a pool of 59 common channels. The proposed channel selection scheme was validated on the UNM dataset and tested on the Iowa dataset to assess its generalizability. The results show that the proposed model achieves an optimal classification accuracy of 100% on the UNM dataset and 93.75% on the Iowa dataset. The channel selection method significantly reduces the dimensionality of EEG feature vectors related to PD by 50%. The study also demonstrates the effectiveness of the proposed method in both the UNM and Iowa datasets, with the highest classification accuracy of 100% for the closed-eye state and 93.75% for the open-eye state. The results indicate that the proposed channel selection approach is effective in improving classification performance and generalization capability. The study also highlights the importance of channel selection in reducing the complexity of EEG data for PD diagnosis. The findings suggest that the proposed method can be applied to other EEG-related tasks, such as epilepsy diagnosis and emotion recognition, to enhance its applicability and utility in various clinical and research settings.