Learning Multi-Class Neural-Network Models from Electroencephalograms

TL;DR

This paper introduces a novel algorithm for training multi-class neural networks on EEG data, effectively handling variability and overlapping classes, with high accuracy and probabilistic decision interpretation.

Contribution

The paper presents a new pairwise training algorithm for multi-class neural networks applied to EEG classification, emphasizing variable relevance and probabilistic outputs.

Findings

Achieved 80.8% training accuracy on EEG data

Achieved 80.1% testing accuracy on EEG data

Provided probabilistic interpretation of neural network decisions

Abstract

We describe a new algorithm for learning multi-class neural-network models from large-scale clinical electroencephalograms (EEGs). This algorithm trains hidden neurons separately to classify all the pairs of classes. To find best pairwise classifiers, our algorithm searches for input variables which are relevant to the classification problem. Despite patient variability and heavily overlapping classes, a 16-class model learnt from EEGs of 65 sleeping newborns correctly classified 80.8% of the training and 80.1% of the testing examples. Additionally, the neural-network model provides a probabilistic interpretation of decisions.