m-granat/cnn-for-eeg-tsc

Electroencephalography (EEG) is an effective and non-invasive way to capture electric activity of the brain. EEG time series classification is a very important problem in neuroscience as a lot of EEG practical applications, such as medical diagnostics and brain–computer interfaces, depend on the quality of classification results. Several effective classification methods that use deep neural networks were developed in recent years, and convolutional neural networks (CNN) have shown exceptional results in many studies. The aim of this project is to investigate if EEG time series can be classified according to the types of attention (mental or sensory) that was used by research participants during the experiment. Deep multi-scale convolutional neural network is used as a classification model. The results have shown that CNN is indeed capable of solving such classification tasks: 98% accuracy was obtained on validation dataset. The obtained results suggest that CNN capabilities in extracting features from time series data is a perspective field for further research.