Brain-Computer Interfaces (BCIs) creates a straight link between the brain and peripheral systems, bypassing the reliance on conventional communication methods. In Brain Computer Interface (BCI) practice, motor imagery (MI) signals recorded through electroencephalography (EEG) are extensively utilized to assist individuals with disabilities, control environments, and enhance human capabilities. Nevertheless, decoding MI EEG data due to its highly non-stationary characteristics and significant variability across subjects remains a complex task. In this study an innovative classification framework is introduced combining Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Pearson Correlation Coefficient (PCC)-based selection, Hilbert-Huang Transform (HHT) feature extraction, and Support Vector Machine (SVM) classification. CEEMDAN is applied to EEG signals decomposing the signal into Intrinsic Mode Functions (IMFs), PCC is then employed to identify the most informative IMFs. Subsequently, temporal and frequency features are extracted through HHT, which are classified using an SVM with a radial basis function (RBF) kernel. The suggested approach is then substantiated upon the BCI Competition IV-2a (BCI-2a) dataset, attaining an accuracy of 88.72%, precision of 89.21%, recall of 88.85%, F1-score of 87.95%, and Cohen’s kappa coefficient of 0.80. These results demonstrate an improvement over conventional classification techniques across multiple performance metrics.