Research & Projects

Projects

Independent research and science-fair work.

Grasp and Lift EEG Detection: Brain-Computer Interface for Hand Movement Recognition Using non-Invasive EEG Data

Project abstract

The objective of this project is to develop a robust classifier for hand movement events from EEG signals, addressing challenges such as inherent noise, low signal-to-noise ratios, and complex multi-channel data. Advanced machine learning techniques and novel feature extraction methods were employed to accurately identify six distinct hand movement events critical for neuroprosthetic control and assistive device operation.

A public dataset of raw EEG signals and six hand movement events from 12 subjects was used; each subject participated in 10 trial series with roughly 30 trials per series. Four models were trained: a Convolutional Neural Network (CNN), a Random Forest classifier, a Random Forest model with extracted and augmented features, and a stacking classifier combining XGBoost with a Multi-Layer Perceptron (MLP). Accuracy was evaluated using 20% of the training data as new data, and the best model was tested on a dataset comprising 2 trial series per subject with about 30 trials each.

The stacking classifier achieved a maximum accuracy of 80% (validated by cross-validation), while the CNN, Random Forest, and augmented Random Forest models yielded accuracies of 45%, 71%, and 78%, respectively. The best model predicted hand movements for the test data.

This work illustrates that non-invasive EEG-based classification can enable reliable brain-computer interfaces for prosthetic control. Future improvements will focus on refining feature extraction and addressing class imbalance, ultimately enhancing motor function restoration and improving quality of life for individuals with neurological disabilities or amputations.

Project materials: coming soon.