Impulsivity Classification Using EEG Power and Explainable Machine Learning.

Area of Science:

• Neuroscience
• Cognitive Science
• Machine Learning

Background:

• Impulsivity is a complex trait linked to negative outcomes.
• Previous research on electroencephalography (EEG) and impulsivity has yielded inconsistent findings.
• A data-driven approach is needed to identify reliable EEG predictors of impulsivity.

Purpose of the Study:

• To identify specific EEG power features for predicting self-reported impulsivity.
• To determine the most effective machine learning classifiers and relevant EEG bands for impulsivity classification.
• To pinpoint the most informative electrode sites and EEG features for accurate impulsivity prediction.

Main Methods:

• Recorded EEG signals from 56 participants during a risk-taking task.
• Utilized machine learning classifiers with EEG power features from 62 electrodes.
• Employed k-fold cross-validation, sequential feature selection, and SHAP analysis for model evaluation and interpretation.

Main Results:

• Alpha and beta band power demonstrated high accuracy (95.18% and 95.11%) in classifying impulsivity using a random forest classifier.
• A subset of 10 electrodes was sufficient for reliable impulsivity classification based on alpha band power (94.50% F1-score).
• SHAP analysis highlighted frontal and posterior midline alpha power as the most significant predictors of impulsivity.

Conclusions:

• EEG alpha and beta band power are robust predictors of impulsivity.
• Machine learning models, particularly random forest, can accurately classify impulsivity levels using EEG data.
• Frontal and posterior midline alpha power are crucial neural correlates of impulsivity, offering potential for future diagnostic tools.