Western Artificial Intelligence

Developer · 2022 — 2023

• Motor Imagery Classification: Developed a custom Convolutional Neural Network (CNN) architecture (3 layers, 128 filters, ReLU) using TensorFlow and Keras to classify EEG signals for Left/Right hand motor imagery.
• Signal Processing Pipeline: Implemented a robust preprocessing pipeline including 4th-order Butterworth band-pass filtering (7.5-30 Hz) for Alpha/Beta waves, Common Spatial Patterns (CSP) for dimensionality reduction, and Stockwell Transform for time-frequency analysis.
• Research & Analysis: Achieved ~60% validation accuracy on the BCI Competition IV 2a dataset. Conducted extensive error analysis to identify overfitting issues and proposed future improvements using Continuous Wavelet Transforms (CWT).
• Conference Presentation: Authored and presented "PiBrain: CNN-based Motor Imagery Classification" at the Canadian Undergraduate Conference on Artificial Intelligence (CUCAI) 2023, detailing the complete BCI pipeline from raw signal to classification.