Enhanced Classification of Motor Imaginary in EEG Using Feature Optimization and Machine Learning

Accurate classification of motor imagery (MI) in EEG signals plays a crucial role in the diagnosis of neurological diseases, including conditions affecting motor control such as brain strokes and amyotrophic lateral sclerosis (ALS). However, the complex and high-dimensional nature of MI-EEG data poses significant challenges for accurate classification. Traditional classification methods often struggle with noise, artifacts, and redundant features, leading to reduced classification accuracy and increased computational complexity.

This paper presents an enhanced classification technique leveraging feature optimization and a deep neural network (DNN) classifier to improve the accuracy of MI-EEG data classification. The proposed approach utilizes a three-layer DNN model integrated with the Teacher Learning-Based Optimization (TLBO) technique. This optimization method reduces noise and artifacts in EEG signals, enhancing the quality of input vectors for the DNN classifier. The feature extraction process employs discrete wavelet transform (DWT) to decompose the EEG signals into multiple sub-bands, capturing essential frequency components. Subsequently, the TLBO algorithm refines these features, optimizing them for improved classification performance.

The proposed algorithm was evaluated using datasets from the third and fourth BCI competitions and simulated within a MATLAB environment. Comparative analysis was conducted against existing algorithms, including Bayesian Networks (BN) and Ensembled Machine Learning (EBL), to validate the performance of the proposed method. Experimental results demonstrate that the suggested approach significantly improves classification accuracy across various EEG signal bands, including raw, delta, theta, alpha, and beta signals. The combination of DNN with TLBO not only enhances classification accuracy but also reduces computational complexity by selecting the most relevant features.

The findings highlight the potential of the proposed approach in developing robust and reliable MI-based Brain-Computer Interface (BCI) applications for motor control, such as assistive communication systems, gaming, and wheelchair control for individuals with motor disabilities. Future work will focus on extending this approach to classify multi-class MI tasks, thereby broadening its applicability in advanced communication and control systems.