MI-BMPI motor imagery brain–mobile phone dataset and performance evaluation of voting ensembles utilizing QPDM

• Published in: Neural computing & applications Vol. 37; no. 6; pp. 4679 - 4696
• Main Authors: Yilmaz, Cagatay Murat, Yilmaz, Bahar Hatipoglu, Kose, Cemal
• Format: Journal Article
• Language: English
• Published: London Springer London 01.02.2025; Springer Nature B.V
• Subjects: Artificial Intelligence; Cell phones; Cellular telephones; Classification; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Data Mining and Knowledge Discovery; Datasets; Electroencephalography; Frequencies; Human-computer interface; Image Processing and Computer Vision; Original Article; Performance evaluation; Probability and Statistics in Computer Science; Voting; Brain–mobile phone interface; Electroencephalography; Motor imagery; Brain–computer interface
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-024-10917-5

EEG-based interfaces are an active research area with great potential. We, therefore, focused on classifying motor imaging (MI) tasks from various problem areas. Because of that, we applied MI patterns to voting ensembles differently and constructed voters. They employ quasi-probabilistic distribution models based on sub-classifiers of different frequency bands and time segments. Much previous work focused on just a few MI tasks for BCIs. To that end, we constructed a new mobile EEG dataset, abbreviated as MI-BMPI, containing two significant gestures for mobile phone interfaces. The research experiments used a consumer market EEG system, the mobile wireless Emotiv EPOC Flex neuroheadset. Experiments were carried out on the BCI Competition IV Dataset 2a and MI-BMPI. On the BCI and BMPI datasets, promising results were obtained in classifying various MI tasks. In conclusion, new solutions were introduced for tougher EEG-based interfaces, which have the potential to classify MI tasks and develop EEG-based interfaces. In addition to the average performance improvements, more stable results were achieved for subject and task variations.