Decoding Motor Imagery (MI) from Electroencephalography (EEG) signals is critical for Brain-Machine Interface (BMI) applications. This study investigates the Redundant Discrete Wavelet Transform (RDWT) to determine the optimal mother wavelet (MW) for enhanced EEG-based MI decoding across diverse frequency bands and topographical regions. The optimal MW selection for EEG analysis using Discrete Wavelet Transform (DWT) remains a challenge. While some studies have addressed MW choice for EEG analysis, few have conducted comprehensive comparative analyses. Existing studies often rely on linear correlation metrics, neglecting residual shifts and boundary effects inherent to DWT. Our investigation leverages EEG data recorded at 100 Hz during longitudinal BMI training with a powered lower-limb exoskeleton. We evaluate magnitude/phase spectra and classification accuracy for ‘No-Go’/‘Go’ tasks. For the assessment of classification accuracy based on MWs, we employ a Neural Network classifier with Time and Frequency Domain (TFD) features extracted using RDWT. Our findings demonstrate that the Neural Network classifier with TFD features extracted using RDWT enhances classification accuracy by 4.5% to 11.9%. Our analysis emphasizes the importance of the shortlisted MWs ‘Symlet2’ and ‘Coiflet1’ across diverse frequency bands, providing valuable insights into EEG signal feature extraction specifically related to MI.