<p>Brain–computer interface (BCI) technology has huge potential to transform human–computer interaction across multiple disciplines. Despite efforts, achieving high accuracy in motor imagery (MI) task classification remains difficult. This proposed solution addresses the MI tasks classification challenge utilizing the BCI IV 2a and 2b datasets. Using optimization methods, optimal frequency bands associated with event-related desynchronization were identified to enhance classification accuracy to 90.44% for 2a and 93.78% for 2b datasets. Following band selection, the electroencephalography data were pre-processed, focusing on the identified frequency band, and quadratic interpolation was employed for data augmentation, introducing synthetic data points to improve model generalization and robustness. The proposed method used wavelet scattering transform for feature extraction, which captures both temporal and spectral characteristics. The extracted features set was reduced using common spatial patterns, which utilizes spatial distribution to find discriminative spatial filters. After feature selection simple classifier, recurrent neural network was implemented and compared with many state-of-the-art methods. The system proposed is computationally efficient and lightweight making it applicable to IoT-based applications. The proposed methodology shows that the technique improves BCI classification performance, providing useful insights for applications in healthcare and assistive technologies.</p>