[Paper Review] Understanding Brain Dynamics for Color Perception using Wearable EEG headband

Mahima Chaudhary, Sumona Mukhopadhyay+3 more|arXiv (Cornell University)|Aug 17, 2020

EEG and Brain-Computer Interfaces26 references8 citations

TL;DR

This study proposes a multiclass classification model using spectral, statistical, and correlation features from raw EEG signals captured by a wearable Muse 2 headband to distinguish brain responses to red, green, and blue color stimuli. By applying forward feature selection and a Random Forest classifier, the method achieved 80.6% intra-subject accuracy, demonstrating that consumer-grade EEG devices can effectively classify color perception for brain-computer interface applications in IoT environments.

ABSTRACT

The perception of color is an important cognitive feature of the human brain. The variety of colors that impinge upon the human eye can trigger changes in brain activity which can be captured using electroencephalography (EEG). In this work, we have designed a multiclass classification model to detect the primary colors from the features of raw EEG signals. In contrast to previous research, our method employs spectral power features, statistical features as well as correlation features from the signal band power obtained from continuous Morlet wavelet transform instead of raw EEG, for the classification task. We have applied dimensionality reduction techniques such as Forward Feature Selection and Stacked Autoencoders to reduce the dimension of data eventually increasing the model's efficiency. Our proposed methodology using Forward Selection and Random Forest Classifier gave the best overall accuracy of 80.6\% for intra-subject classification. Our approach shows promise in developing techniques for cognitive tasks using color cues such as controlling Internet of Thing (IoT) devices by looking at primary colors for individuals with restricted motor abilities.

Motivation & Objective

• To develop a low-cost, wearable BCI system that classifies brain responses to primary colors using consumer-grade EEG hardware.
• To investigate whether spectral, statistical, and correlation features from wavelet-transformed EEG signals improve classification accuracy over raw signals.
• To evaluate the effectiveness of dimensionality reduction techniques like forward feature selection and stacked autoencoders in enhancing model efficiency and performance.
• To compare intra-subject and inter-subject classification performance for color perception using machine learning models.
• To enable practical applications such as IoT control for individuals with motor impairments using color-based brain signals.

Proposed method

• Collected raw EEG signals from a four-channel Muse 2 wearable headband during exposure to red, green, and blue color stimuli.
• Preprocessed signals using continuous Morlet wavelet transform to extract band power in alpha (8–13 Hz) and beta (13–30 Hz) frequency bands.
• Extracted spectral power, variance in power, pairwise cross-correlation features, and statistical features (e.g., mean, kurtosis) from wavelet-transformed band power.
• Applied forward feature selection (supervised) and stacked autoencoders (unsupervised) for dimensionality reduction to improve model efficiency.
• Trained and evaluated multiple machine learning models including KNN, SVM, Logistic Regression, Random Forest, Neural Networks, and Gradient Boosting on the reduced feature set.
• Optimized model performance using a 200ms time window to isolate transient responses to color stimuli and minimize motion artifacts.

Experimental results

Research questions

• RQ1 Is it possible to distinguish EEG signals from a four-channel wearable headband, produced by RGB color exposure, by training ML models on features that account for statistical, spectral and correlation properties of EEG?
• RQ2 Can feature reduction techniques like Forward-Feature Selection and Autoencoders make the ML algorithms for EEG classification more efficient?
• RQ3 Does the performance of ML algorithms differ for inter-subject and intra-subject classification?
• RQ4 Can a consumer-grade wearable EEG device achieve high classification accuracy for color perception using advanced feature engineering and ML?

Key findings

• The proposed model using Random Forest with forward feature selection achieved the highest intra-subject classification accuracy of 80.6%.
• The ROC-AUC score for the best subject reached 0.730 using the Gradient Boost model with autoencoder-processed features, indicating strong discriminative performance.
• Inter-subject classification accuracy was lower at 39.8%, suggesting significant inter-individual variability in EEG responses to color stimuli.
• Forward feature selection outperformed stacked autoencoders in improving model accuracy and efficiency, particularly for intra-subject classification.
• The Random Forest model with forward selection achieved a balanced F1-score of 0.510 and MCC of 0.225 on average across subjects, indicating moderate to good performance.
• The study demonstrated that a 4-channel consumer-grade EEG device (Muse 2) can achieve comparable performance to higher-electrode medical systems, with 80.6% accuracy—surpassing previous 4-channel studies (58%) and approaching results from 14-channel systems (79.6%)