Using a Novel LDA-Ensemble Framework to Classification of Motor Imagery Tasks for Brain-Computer Interface Applications