Metaheuristic Optimization-Based Feature Selection for Imagery and Arithmetic Tasks: An fNIRS Study.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Brain-computer interfaces (BCIs) are crucial for assisting individuals with neurological impairments.
• Feature selection is essential for optimizing BCI performance by reducing data dimensionality and improving computational efficiency.
• Functional near-infrared spectroscopy (fNIRS) offers a non-invasive method for monitoring brain activity, making it suitable for BCI applications.

Purpose of the Study:

• To develop and evaluate a wrapper-based metaheuristic feature selection framework for fNIRS-based BCIs.
• To assess the effectiveness of seven metaheuristic optimization algorithms in identifying optimal features for BCI tasks.
• To improve classification accuracy and reduce feature vector size in fNIRS-BCI systems.

Main Methods:

• Temporal statistical features (mean, slope, maximum, skewness, kurtosis) were extracted from fNIRS data.
• Seven metaheuristic algorithms (PSO, CS, FA, BA, FPA, WOA, GWO) were employed for feature selection.
• A k-nearest neighbor classifier was used to evaluate the performance of selected features on motor imagery and mental arithmetic tasks.

Main Results:

• Metaheuristic feature selection significantly improved classification accuracy compared to using the full feature set.
• All tested algorithms reduced the feature vector size while enhancing classification performance.
• Grey Wolf Optimization (GWO) achieved the highest average classification rates: 94.83% for MA, 92.57% for MI, and 85.66% for a four-class task.

Conclusions:

• The proposed metaheuristic feature selection framework effectively enhances the performance of fNIRS-based BCIs.
• GWO demonstrated superior performance in selecting discriminative features for motor imagery and mental arithmetic tasks.
• This framework can aid in the training phase for developing robust and accurate fNIRS-BCI applications.