fNIRS Based Comparative Study of Classifiers and Feature Selection Techniques for Finger Tapping.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Functional near-infrared spectroscopy (fNIRS) is a non-invasive neuroimaging technique.
• Classifying fine motor movements is crucial for brain-computer interface (BCI) development.
• Existing machine learning (ML) models require optimized feature sets for improved performance.

Purpose of the Study:

• To classify five-finger movements using ML algorithms.
• To evaluate the impact of feature optimization on classification performance.
• To identify optimal feature selection methods for fNIRS-based BCI applications.

Main Methods:

• Acquired fNIRS signals from 20 participants performing five distinct finger movements.
• Extracted 17 spatial features from the fNIRS signals.
• Employed Support Vector Machine (SVM) and Extreme Gradient Boosting (XGBoost) classifiers.
• Utilized Genetic Algorithm (GA), Ant Colony Optimization (ACO), and Particle Swarm Optimization (PSO) for feature selection.

Main Results:

• Feature optimization significantly improved ML classification performance.
• GA and PSO outperformed ACO in feature selection.
• XGBoost surpassed SVM in classification accuracy.
• The highest accuracy achieved was 94.94% using GA-optimized features with XGBoost.

Conclusions:

• Feature selection is vital for enhancing the efficiency and accuracy of ML models in neuroimaging.
• Optimized classification pipelines can improve BCI system performance.
• This study provides a framework for effective fNIRS signal classification for BCI applications.