Jämförelse av prestandan hos traditionella maskininlärningsmodeller och djupinlärningsmodeller i realtidsklassificering inom hjärn-datorgränssnitt

Brain-Computer Interfaces (BCIs) are systems that enable direct communication between the human brain and external devices, offering promising applications for individuals with motor impairments. To function effectively, these systems must be able to translate brain signals, such as those captured by electroencephalography (EEG), into reliable commands in real-time. This requires classification models that balance high predictive accuracy with low computational overhead. This study compares the performance of traditional machine learning models and deep learning models for real-time EEG classification in BCI systems. Two models from each category were implemented: shrinkage Linear Discriminant Analysis (sLDA) and Riemannian Minimum Distance to Mean (RMDM) as traditional models, and EEGNet and ShallowConvNet as deep learning models. All models were trained and evaluated on a public motor imagery dataset from BCI Competition IV in offline and simulated online conditions. Performance was assessed using accuracy, precision, recall, F1-score, and processing time. The results show that both model categories perform similarly across the performance metrics. Importantly, a substantial difference is observed in processing time: traditional models, especially sLDA, perform up to 250 times faster than deep learning models. These findings suggest that while deep learning models perform competitively in controlled settings, traditional models currently provide a more practical and efficient solution for real-time BCI applications.