Novel Feature Generation for Classification of Motor Activity from Functional Near-Infrared Spectroscopy Signals...

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Cognitive motor action decoding relies on analyzing Functional Near-Infrared Spectroscopy (fNIRS) data.
• Accurate detection in non-stationary fNIRS signals is challenging due to complex feature requirements.

Purpose of the Study:

• To develop a novel framework for enhancing the accuracy of classifying mental drawing (MD) and spatial navigation (SN) using fNIRS data.
• To introduce a new fused feature by combining wavelet, Hilbert, symlet, and Hjorth parameters.

Main Methods:

• Implemented Independent Component Analysis (FastICA, Picard, Infomax) for noise reduction.
• Developed two binary classifiers for MD and SN detection.
• Utilized Light Gradient-Boosting Machine (LGBM) and Extreme Gradient Boosting (XGBOOST) algorithms.

Main Results:

• The proposed fused feature significantly improved classification accuracy.
• Light Gradient-Boosting Machine (LGBM) achieved 98% accuracy for mental drawing and 97% for spatial navigation.
• Statistical validation confirmed the reliability of the new feature generation method.

Conclusions:

• The novel fused feature framework offers superior performance for cognitive motor action decoding from fNIRS.
• The proposed method surpasses existing approaches in classification accuracy.
• This research provides a reliable mechanism for accurate fNIRS signal analysis.