Feasibility of local interpretable model-agnostic explanations (LIME) algorithm as an effective and interpretable...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Functional near-infrared spectroscopy (fNIRS) is a widely used neuroimaging technique.
• Various feature selection methods exist for fNIRS data, but Local Interpretable Model-Agnostic Explanations (LIME) lacks validation.
• There is a growing need to assess LIME's efficacy in fNIRS studies.

Purpose of the Study:

• To evaluate the feature selection performance of LIME for fNIRS datasets.
• To compare LIME against benchmark, filter-based, and wrapper-based feature selection methods.
• To determine if LIME enhances classification accuracy in fNIRS data.

Main Methods:

• Comparative analysis of feature selection techniques on open-access fNIRS datasets.
• Inclusion of LIME, Minimum-Redundancy Maximum-Relevance (mRMR), t-test, Sequential Forward Selection (SFS), and a no-feature-selection benchmark.
• Evaluation based on classification accuracy.

Main Results:

• LIME demonstrated superior performance compared to other feature selection methods in most evaluated fNIRS datasets.
• LIME achieved statistically significant improvements in classification accuracy over the benchmark.
• The performance of LIME surpassed filter-based and wrapper-based methods.

Conclusions:

• LIME is an effective feature selection method for fNIRS datasets.
• The validated performance of LIME offers a promising approach for enhancing fNIRS data analysis and interpretation.
• LIME's ability to improve classification accuracy highlights its potential in brain-computer interfaces and clinical applications using fNIRS.