Enhancing cognitive state detection through deep Forest-based electroencephalogram signal analysis and classification

The impacts of cognitive tasks on the brain through Electroencephalogram (EEG) signal analysis have commonly employed machine learning models like Support Vector Machines (SVMs), Artificial Neural Networks (ANNs), random forests etc. However, these traditional models may encounter limitations in effectively addressing the unique challenges inherent in EEG signal analysis, including high dimensionality and the potential presence of noise and artefacts. This critique underscores the need for advanced methodologies, capable of navigating these challenges to enhance the accuracy and reliability of cognitive task-related EEG studies. This study addresses the challenge of accurately detecting cognitive states using EEG signals. This helps in overcoming the challenges by high dimensionality, non-stationarity, and noise in raw EEG data. Traditional classifiers such as SVMs and ANNs often fail to fully exploit the temporal and frequency features present in EEG. In order to overcome these limitations, this work introduces a novel Deep Forest-based classification model. This model is optimised through XGBoost for feature selection. A self-acquired dataset using the g.Nautilus EEG device from 12 student volunteers performing cognitive tasks forms the basis of the evaluation. A total of 48 features, spanning time, frequency, entropy, and autoregressive domains, are extracted. The proposed model achieves high classification accuracy (99.692%) while reducing computational time. This method demonstrates strong potential for real-time cognitive monitoring in neuroscience and human-computer interaction contexts.