EEG Multi-Objective Feature Selection using a Genetic Procedure with Hybrid Mutation Operator
Corina Cîmpanu · 2023
Computer science developments in medical data processing increase modeling accuracy, speed up data analysis, and obtain automated diagnostics. As a result, the role of machine learning algorithms becomes significant, as well as the need to implement faster and more accurate systems. For example, classification is a common component in computer-aided diagnostics, and choosing the proper set of predictors is critical for training a classifier model. Usually, medical datasets are highly dimensional, and their characteristical data correlations require reducing the set of features used for classification. Besides supervised grouping algorithms, genetic optimization procedures are well known for their robust verification of competing attributes by directly estimating their usefulness on a specific classifier. This paper discusses the assessment of Electroencephalogram data samples acquired during working memory load n-back scenarios using a new embedded feature selection procedure based on Multi-Objective Optimization via Genetic Algorithms enhanced with a hybrid mutation operator. Experimental results focus on both the hybrid mutation operator performances, and obtaining overall higher Support Vector Machine classification accuracy while reducing the number of selected features.