HYP-PSO-XGB: efficient hyperparameter tuning of XGBoost for EEG-based hand gesture classification using Particle Swarm Optimization

A very crucial pursuit in Brain-Computer Interface technology is to achieve highly reliable classification of hand gestures with the use of Electroencephalogram (EEG) data. It is momentous to enrich the communication of audio-vocally challenged individuals. This paper proposes a novel method for the calibration of the hyperparameters of the Extreme Gradient Boosting (XGBoost) Classifier, dependent on the data complexity, through Particle Swarm Optimization (PSO), aiming to increase the overall sensibility of the model for the hand movements of different kinds, based on the EEG. In this approach, several crucial hyperparameters were adjusted multiple times based on the best trade-offs of the other hyperparameters. The datasets involving EEG data for monitoring distinct hand movements were preprocessed prior to being divided into training and testing datasets. The use of Particle Swarm Optimization (PSO) enabled a more flexible and several iterations quality-adjustment of the hyperparameters, with the eventual goal of boosting the classification accuracy. The new PSO-optimized XGBoost model reached an accuracy of 93.57%. This value represents a substantial increase over existing models, i.e. Improved Particle Swarm Optimization model with 76.67%, PSO SVM model for Motor Imagery with 92.92%, PSO SVM model with 80.63%, and GAPSO-SVM model with 88.89%. It proves the potential of PSO to significantly improve the efficiency of XGBoost, making it possible to create even more sensitive and nuanced BCIs aiding hand gesture classification.