Efficient Feature Selection Technique based on Fast Fourier Transform with PSO-GA for Functional Magnetic Resonance Imaging

Advent of Functional Magnetic Resonance Imaging (fMRI) has helped researchers to know about the dynamic states of brain. One of the maj or goal of fMRI research is the task classification based on the activity of brain states. Every time, it remains a challenge for such researchers to extract optimum features from brain images due to high dimensional quantity of the data in fMRI Images. In this paper, the authors proposed a novel feature selection technique based on Fast Fourier Transform with Particle Swarm Optimization and Genetic Algorithm (FFT-PSOGA) for extracting the best features in fMRI dataset. The resulted features are trained for machine learning models of GaussianNB, Support Vector Machine, and XGBoost and compared for performance with state-of-art. The achieved results are outperforming the existing works based on the same dataset. The authors believe that the feature selection technique proposed in this paper will be the optimal method for extracting brain images' features in any fMRI dataset which will improve the classification accuracy of the decoding of brain images in a much better way.