Brain–computer interfaces (BCIs) have increasingly focused on decoding fine motor movements, particularly hand gestures, to enhance neurorehabilitation and assistive technologies. This study explores the neural representation of coordinated five-finger movements using an electroencephalography (EEG) dataset from four subjects. A multi-stage preprocessing pipeline was employed to improve EEG signal quality, incorporating moving average filtering, common average referencing, and band-pass filtering. This combination of techniques was selected as it effectively enhances signal quality and minimizes noise, leading to improved feature extraction performance. A Wasserstein Generative Adversarial Network (WGAN) was employed for feature extraction, utilizing a discriminator to generate highly discriminative neural features. These features were then classified using various machine learning models, with the Support Vector Machine (SVM) achieving an average classification accuracy of 95.98%, and an average F1-score of 96.00%. This highlights the effectiveness of adversarial feature learning in enhancing neural feature extraction for precise motor intention decoding. By demonstrating a high-accuracy, subject-specific approach to motor control, this study paves the way for incorporating BCIs into rehabilitation systems aimed at restoring fine finger movements in stroke patients.