<p>Motor imagery (MI) signals derived from Electroencephalogram (EEG) provide the most practical foundation for developing Brain-Computer interfaces (BCIs). Researchers face challenges in extracting and selecting relevant features from EEG signals that improve classifier performance in BCI systems. The WPD technique up to the 8th level is used to extract crucial features and calculate Energy, Standard Deviation, Variance, Kurtosis, Skewness, and Approximate Entropy from MI-EEG signals. This research work presents a novel technique for classifying MI signals in BCIs by combining wavelet packet decomposition (WPD) for feature extraction process and feature selection with a Modified Whale Optimization Algorithm (MWOA) integrated with Differential Evolution (DE). The suggested MWOA maintains an appropriate balance between exploration and exploitation, limiting premature convergence and handling the immense dimensionality of MI data. The Support Vector Machine (SVM) is used for classification that employs three kernel functions: the radial basis function (RBF) kernel, the linear kernel, and the sigmoid kernel. The proposed model shows classification accuracy using SVM with an RBF of 95.44%, a precision of 88.49%, a sensitivity of 84.29%, and an <i>F</i>1 score of 80.64% on the benchmark BCI IV-dataset I. Experimental outcomes demonstrate that the WPD + MWOA + SVM framework beats the previously published research on the same dataset and improves performance for real-time BCI applications. This integrated technique permits more effective control in neuro-prosthetics and rehabilitation, increasing the utility of BCIs in real-time applications.</p>