A hybrid supervised learning model for the classification of a low-volume EEG dataset employing powerful ensembles
Abstract
Training Electroencephalogram (EEG)-based detection models with limited observations to achieve high test accuracy is challenging. This work facilitates a straightforward design that expands sample size through linear signal transformation and a convolutional neural network (CNN) and avoids fitting issues with a hybrid Stacking Classifier (SC). The specialized layers of the CNN extract meaningful features from the EEG using limited max and average pooling, dropout, and convolutional layers, thereby building a less complex CNN. The SC framework is a novel approach that uses heterogeneous base-ensemble models arranged in a stack to learn data patterns for efficient classification. This model leverages the strengths of various single machine learning models, as well as bagging and boosting ensembles, to manage the bias-variance trade-off, which is especially evident when working with hybrid models on a limited dataset. The CNN-SC model learns patterns using base learners, aggregates results via a meta-ensemble layer, and achieves an accuracy of 0.957 and a sensitivity of 0.97. The challenges encountered during the development of hybrid deep and machine learning algorithms are carefully addressed, resulting in low test error rates and processing time. This powerful model serves as an inspiration for healthcare diagnostic applications with low-volume datasets that face data-collection challenges.
Related
- Text guidance is powerful but prompt-sensitive for weakly-supervised leaf symptom segmentation
- An urban patrol case urgency classification method based on self-supervised clustering of a graph neural network
- Learning from Error Predictions Rather Than Prediction Errors: A Perspective on Self-Supervised Learning in the Brain
- Exploring noise in TCR specificity data through supervised learning
- Metabolite-centric Knowledge Graph Unveils Predictive and Mechanistic Insights through Self-supervised Multimodal Learning
BibTeX
@article{mulamoottil2026hybrid,
title = {{A hybrid supervised learning model for the classification of a low-volume EEG dataset employing powerful ensembles}},
author = {Steffi Philip Mulamoottil and T. Vigneswaran},
journal = {Scientific Reports},
year = {2026},
month = sep,
doi = {10.1038/s41598-026-65131-1},
url = {https://www.nature.com/articles/s41598-026-65131-1},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - A hybrid supervised learning model for the classification of a low-volume EEG dataset employing powerful ensembles
AU - Steffi Philip Mulamoottil
AU - T. Vigneswaran
PY - 2026
DA - 2026/09/30
JO - Scientific Reports
DO - 10.1038/s41598-026-65131-1
UR - https://www.nature.com/articles/s41598-026-65131-1
AB - Training Electroencephalogram (EEG)-based detection models with limited observations to achieve high test accuracy is challenging. This work facilitates a straightforward design that expands sample size through linear signal transformation and a convolutional neural network (CNN) and avoids fitting issues with a hybrid Stacking Classifier (SC). The specialized layers of the CNN extract meaningful features from the EEG using limited max and average pooling, dropout, and convolutional layers, thereby building a less complex CNN. The SC framework is a novel approach that uses heterogeneous base-ensemble models arranged in a stack to learn data patterns for efficient classification. This model leverages the strengths of various single machine learning models, as well as bagging and boosting ensembles, to manage the bias-variance trade-off, which is especially evident when working with hybrid models on a limited dataset. The CNN-SC model learns patterns using base learners, aggregates results via a meta-ensemble layer, and achieves an accuracy of 0.957 and a sensitivity of 0.97. The challenges encountered during the development of hybrid deep and machine learning algorithms are carefully addressed, resulting in low test error rates and processing time. This powerful model serves as an inspiration for healthcare diagnostic applications with low-volume datasets that face data-collection challenges.
SN - 2045-2322
ER -