Deep Learning Identifies Brain Cognitive Load Via EEG Signals

The process of EEG signal analysis involves multi-frequency non-stationary brain waves from many channels. Appropriate segmentation of these signals, extracting discriminatory features to obtain the imperative properties, and classification are the key aspects of a pattern detection algorithm. The performance of the machine learning models is highly dependent on feature extraction and selection techniques. Identification of best possible features requires intricate domain knowledge and cumbersome exploration. To evade such dependencies, deep learning methods have evolved for EEG detection, which automatically extract the best features needed for the classification. However, the time to train such networks is substantial due to the complex structure of the model. To address these issues, the present paper proposes an automatic system for brain cognitive state detection using a simple 12-layer 1D network architecture. The developed algorithm yields ceiling level classification performance for both the experimental and the benchmark dataset.