Higher-Order Dynamic Mode Decomposition and Multichannel Singular Spectrum Decomposition Hybridization for BCI Feature Extraction

A Brain-Computer Interface (BCI) communicates between the human brain and the different external intelligent devices. The main objective of the BCI system is to translate human intentions into computer-based control commands. This paper demonstrates that a hybrid feature engineering approach is more efficient in characterizing human intentions than features extracted from single-mode algorithms. We used a randomized amalgamation process with variance maximization to merge features obtained from two recently developed feature extraction techniques: (1) Multichannel singular spectrum decomposition and (2) Higher-order dynamic mode decomposition. The Extreme Learning Machine (ELM) classifier is used with a five-fold cross-validation technique to discriminate multiclass Motor Imagery (MI) patterns. The proposed model achieves a superior Classification Accuracy (CA) of 80.66% compared to the state-of-the-art classification models. Additionally, our method obtains the highest kappa score, which is more robust than existing classification models.