EEG/fNIRS Cognitive State Classification

Research Overview

Complete EEG-based emotional state classification system using real human data from the DEAP dataset. Detects emotional arousal levels (low vs high excitement) from brain signals with 62.5% accuracy using 256 EEG features.

• Real Human Data — 100% DEAP dataset (no synthetic data)
• Binary Classification — Arousal detection (low vs high excitement)
• Feature Engineering — 256 EEG features (32 channels × 8 metrics)
• Machine Learning — Random Forest with comprehensive evaluation
• Neurophysiological Insights — Channel and frequency band analysis

Technical Implementation

End-to-end pipeline from raw EEG signals to emotional state predictions with comprehensive feature analysis and model evaluation.

• Data Processing — DEAP dataset integration with 2 subjects, 80 trials
• Feature Extraction — Beta, theta, alpha bands + statistical metrics
• Classification — Random Forest with cross-validation
• Performance Analysis — Feature importance, confusion matrix, ROC curves
• Optimization — SMOTE balancing and PCA dimensionality reduction

Key Results

Achieved above-chance classification accuracy with comprehensive neurophysiological insights and feature analysis.

• Base Performance — 62.5% accuracy with 256 features
• Balanced Dataset — 68.2% accuracy with SMOTE + 15 features
• PCA Optimized — 72.4% accuracy with dimensionality reduction
• Top Features — Central beta band activity (ch12_beta most predictive)
• Brain Regions — Motor cortex, frontal, and parietal areas most important

Research Paper

Contact me for the full research paper and complete implementation details.