A Generalizable Efficient Machine Learning Framework for Schizophrenia Classification Using Multiscale EEG Features and Ensemble Methods

EEG-based automated classification pipelines for identifying mental disorders increasingly rely on deep learning architectures that are computationally intensive and difficult to interpret, limiting reproducibility and clinical deployment in resource-constrained or cross-site settings. There is a need for algorithmically transparent frameworks that balance accuracy, generalization, and computational efficiency. We propose an interpretable EEG classification framework that integrates multiscale spectrotemporal feature extraction with ensemble machine learning. The pipeline combines standardized preprocessing with extraction of time-domain, spectral power, and entropy-based features, followed by minimum redundancy–maximum relevance feature selection. Classification is performed using voting and stacking ensembles of heterogeneous base learners. The proposed algorithm achieved 98.06% accuracy on a primary open EEG dataset (Warsaw IPN; 19 channels, 250 Hz) and 91.47% accuracy on an independent external dataset (Moscow adolescent cohort; 16 channels, 128 Hz) without retraining or dataset-specific tuning. The framework exhibited low computational overhead and stable cross-dataset performance. The results demonstrate a generalizable, computationally efficient, and interpretable EEG classification framework that favors feature-level transparency and ensemble diversity over deep architectures, supporting scalable and reproducible biomedical signal processing applications.