Deep Forest based EEG Signal Analysis and Classification

This work focuses on the development of a deep forest-based EEG signal analysis and classification system for the detection of cognitive states. The dataset used in this study was self-acquired using a g. Nautilus EEG recording device with six student volunteers focusing on six cognitive tasks. The dataset was preprocessed and feature engineered using various signal processing techniques. The system utilises various features extracted from EEG signals, including Hjorth parameters, wavelet trans-form features, and autoregressive models. The feature dataset is then trained on a cascade forest classifier using hyper-parameter tuning with the XGBoost library to improve the accuracy of the classification model. The top 20 most important features which the classifier selected are identified, and its significance is discussed. The proposed system shows promising results, achieving an overall accuracy of ${9 9. 7 2 1 \%}$ in classifying cognitive states from EEG signals. The proposed method can be used in various fields, such as neuroscience, healthcare, and human-computer interaction, to analyse and monitor cognitive states in real-time