Improving classification performance in hybrid (EEG and NIRS) brain computer interface systems

A brain-Computer Interface (BCI) is a system that enables the signals obtained from neural activities in a person's brain to be processed as commands using a computer system. A BCI system consists of a user, a computer, and peripherals. EEG and NIRS are the primary imaging systems for representative brain signals. The performance of BCI systems is directly proportional to the classifier's performance. Hybrid systems are used to overcome the limitations of one-sided systems and increase the accuracy of the classifier. In this study, after feature extractions were made on a data set that is open to use, the Multiple Instance Learning method was applied, and the performances of various classifiers were measured. Feature extraction operations were performed in the time domain and frequency domain. Naive Bayes, Random Forests, k-Nearest Neighbor, Linear Discriminant Analysis, and Support Vector Machines, the most well-known classification algorithms, are classifiers. The fusion at the attribute level positively affected the performance. The binary classification performance obtained from the EEG-DEOXY hybrid system was 0.985 with the k-Nearest Neighbor algorithm. The Multiple-Instance Learning algorithms positively affected the classification performances.