Explaining electroencephalogram channel and subband sensitivity for alcoholism detection.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alcoholism is a progressive disorder impacting mental and physical health.
• Early detection of alcoholism is crucial for timely intervention and management.
• Electroencephalogram (EEG) signals offer a non-invasive method for brain activity monitoring.

Purpose of the Study:

• To develop an automated system for alcoholism detection using EEG signals.
• To identify reliable biomarkers for alcoholism through explainable artificial intelligence (XAI).
• To evaluate the performance of machine learning models and oversampling techniques in EEG-based alcoholism detection.

Main Methods:

• EEG signals were decomposed into five frequency bands.
• Machine learning models, including Artificial Neural Network (ANN), were trained and evaluated.
• Three oversampling techniques (SMOTE, ADASYN, Gaussian) were used to enhance model generalization.
• Explainable AI techniques (LIME, Submodular Pick LIME, Morris sensitivity analysis) were applied to interpret model predictions.

Main Results:

• ANN achieved superior performance with 97.36% accuracy and 97.88% F1-score.
• SMOTE-based ANN further improved performance to 97.93% accuracy and 97.99% F1-score.
• XAI identified beta and gamma bands, along with parietal and central EEG regions, as significant biomarkers.

Conclusions:

• ANN models effectively detect alcoholism using EEG signals.
• XAI provides valuable insights into the neural correlates of alcoholism.
• The identified biomarkers and methods hold potential for early alcoholism detection and monitoring systems.