Reza Dianat

The brain-computer interface (BCI) is a new field in engineering and medical sciences that focuses on the investigation of controlling external devices through the analysis of brain signals. Recording brain signals using the electroencephalography (EEG) method is a conventional approach in BCI design due to its direct registration of neuronal activity, simplicity of the recording process, and low cost. The objective of this project is to classify recorded brain activities during a cognitive task and automatically detect mental activity/rest using multi-channel EEG signals to understand the relationship between cognitive processes and brain responses, and to provide a BCI. The results of this classification can be used in the design of binary BCIs. At first, preprocessing was performed on the data obtained from a number of volunteers. Then, these data were segmented, and nonlinear statistical features, including various entropies from each segment, were extracted. By selecting an appropriate classifier, each segment was classified into two states, rest and computational activity, in both monopolar and bipolar modes, and evaluated using three different evaluation methods. The best accuracy in the holdout, k-fold, and LOOCV methods was reported as 97%, 94%, and 94%, respectively. Based on the proposed classification and feature selection method, it was determined that the random forest classifier outperformed other tested classifiers and provided higher accuracy. Additionally, bipolar montage showed better performance compared to monopolar montage, with this higher accuracy being attributed to the reduction of common noise in bipolar montage compared to monopolar montage.