EEG Feature Engineering for Motor Imagery Classification Using Efficient Machine Learning Approach

Feature engineering is the core problem in pattern recognition of electroencephalogram signals, which directly affects the design and performance of classifiers. Feature engineering involves transforming the original high-dimensional original signal space into a low-dimensional feature space through some transformations. This process reveals the features that are not easy to be observed and detected in the original feature domain in the transformed domain for making the differences between different categories of features larger, so as to provide the optimal input for the classifier and improve the accuracy of pattern recognition. In this paper, a feature extraction based on wavelet decomposition and subspace energy, as well as a feature selection method based on gene optimization algorithm are proposed, and an ensemble learning algorithm based on optimized probabilistic neural networks is used for classification. Moreover, the BCI Competition 2005 Dataset IIIb is used as the training data to verify the effectiveness and superiority of the proposed algorithm by comparing with traditional feature extraction algorithms.