NeuroFeat: An adaptive neurological EEG feature engineering approach for improved classification of major depressive disorder

Abstract

Major depressive disorder (MDD) can greatly affect individuals’ physical and mental health and can lead to weakened functioning and increased suicide risk. Early diagnosis is crucial for effective treatment, improving outcomes, and reducing long-term societal and healthcare burdens. This research presents an adaptive feature engineering framework for electroencephalogram (EEG) data aimed at enhancing the diagnosis of MDD. The proposed method leverages NeuroFeat, an innovative approach to extracting features from EEG signals. This involves deriving textural features through statistical thresholds and moment analysis from signals decomposed via the discrete wavelet transform. To further optimize feature selection, a logarithmic-spatial bound whale optimization algorithm is employed to refine the extracted featurespace. Thirty different classifiers, including both classical machine learning and deep learning techniques, are utilized to evaluate the performance of the engineered features. Experimental results demonstrate the effectiveness of the featurespace, achieving a binary classification accuracy of up to 99.22% with Cosine k-nearest neighbors and an average accuracy of 88% across all classifiers. This study highlights the effectiveness of advanced feature engineering techniques in EEG-based diagnostic tools, offering a non-invasive and efficient method for the early detection and accurate identification of MDD. • An efficient adaptive feature engineering methodology is developed, namely, NeuroFeat. • NeuroFeat extracts features from EEG signals used for depressive disorder diagnosis. • A novel logarithmic-spatial bound whale optimization algorithm (L-SBWOA) is proposed. • L-SBWOA can select the most relevant features out of the primary feature space. • Thirty different machine/deep learning models are utilized for performance evaluation.

Topics & Concepts