EEG CHANNEL SELECTION USING DIFFERANTIAL EVOLUTION ALGORITHM AND PARTICLE SWARM OPTIMIZATION FOR CLASSIFICATION OF ODORANT-STIMULATED RECORDS

Mesut Şeker,Mehmet Siraç Özerdem

doi:10.36222/ejt.873351

Abstract

A significant advancement has been made in the evolutionary computing and swarm intelligence methods in past decades. These methods have been commonly used to calculate well optimized solutions. Methods select the best elements or cases among set of alternatives. In EEG signal processing applications, efficient channel selection algorithms are required to reduce high dimensionality and remove redundant features. To do this, we examined optimal 5 electrodes out of 14 using Particle Swarm Optimization (PSO) and Differential Evolution Algorithm (DEA). The proposed work is related with pleasant-unpleasant EEG odors classification problem. Classification error rates were calculated by Linear Discriminant Analysis (LDA), k-NN (k Nearest Neighbor), Naive Bayes (NB), Regression Tree (RegTree) classifiers and used as fitness function for optimization algorithms. The results showed that PSO with selected 5 channels gave lowest error rates compared with DEA for all runs. RegTree classifier generated optimal fitness function value among other classifiers. PSO algorithm can effectively support channel selection problem to identify the best channels to maximize classification performance.

Highlights

EEG signals have commonly used in many applications such as motor imagery, mental task, and sleep stage classifications in addition to emotion recognition, seizure detection and drug effects diagnosis
We examined optimal 5 electrodes out of 14 using Particle Swarm Optimization (PSO) and Differential Evolution Algorithm (DEA)
The results showed that PSO with selected 5 channels gave lowest error rates compared with DEA for all runs

Summary

Introduction

EEG signals have commonly used in many applications such as motor imagery, mental task, and sleep stage classifications in addition to emotion recognition, seizure detection and drug effects diagnosis. Efficient channel selection algorithms are required to remove redundant contents from EEG signals. Wavelet transform, and power spectral estimation can be considered as signal processing tools for feature extraction and channel selection algorithms [2]. Population based search procedures such as Particle Swarm Optimization (PSO) and Differential Evolution Algorithm (DEA) have been in demand for researchers in past decades [3]. These algorithms look for the best subset of channels by individually assessing the usefulness of each channel with the help of search engine and fitness function

Methods

Results

Conclusion