Multi-Objective Evolutionary Design of Explainable EEG Classifier

Deep neural networks (DNNs) have achieved impressive results in many fields.
However, the use of black-box solutions based on DNNs in medical applications
poses challenges, as understanding the rationale behind decisions is crucial for
application in healthcare. For those reasons, we propose a new method for the
evolutionary multi-objective design (MOD) of small and potentially explainable
EEG (Electroencephalography) signal classifiers. We evaluate a combination of
genetic algorithm (GA) for feature selection with multiple algorithms for the
automated design of the classifier, including Support Vector Machine, k-Nearest
Neighbors, and Naive Bayes. To further improve the classification quality and
obtain less complex solutions, we compare three different MOD scenarios targeting
the accuracy, specificity, sensitivity, and the number of used features. In
addition, we evaluate the use of Cartesian Genetic Programming (CGP) as a way to
achieve smaller and more interpretable solutions and combine it with the
compositional co-evolution of selected features to improve computational
requirements and find solutions in a reasonable time. The proposed methods are
experimentally evaluated on tasks of alcohol use disorder and major depressive
disorder classification. Experimental results show that newly proposed MOD
scenarios lead to significantly better trade-offs between the accuracy and the
number of features compared to the state-of-the-art method employing the NSGA-II
algorithm. The proposed co-evolution of features (evolved by GA) and classifier
(evolved by CGP) allowed the design of small and potentially explainable
solutions and led to 20-100 times faster convergence than the baseline CGP-based
approach.