AI Research Project - Decoding Human Movement from Brain Signals

Outcome- This research will help to enable brain-controlled prosthetic limbs, wheelchairs, and assistive devices for individuals with paralysis, spinal cord injuries, or limb loss when physical motion is impaired.

Tech Stack - MATLAB

GitHub - https://github.com/swrraab/Decoding-Human-Movement-from-Brain-Signals

Publication - Submitted my research paper for peer review to Curieux Academic Journal and National High School Journal of Science (NHSJS). Currently awaiting editorial decisions.

Overview:

• Developed machine learning models that can decode brain signals using intended movement direction from neural signals recorded in the motor cortex.

• This began when I started exploring research in brain-computer interfaces (BCIs), particularly in prosthetics. While studying this field, I identified a major gap: how many people have tested models for this specific cause, yet no one has directly compared the performance of each model.

• My father helped me with this subject of interest and got me introduced to a PhD student of biomedical engineering from the University of Chicago to understand the depth of this topic and its applications in MATLAB.

• To address this gap, I designed a study where I used MATLAB to compare the Random Forest (RF) and K-Nearest Neighbors (KNN) algorithms to decode the intended movement direction from motor cortex activity.

• I analyzed and collected spike train data from 143 directionally tuned neurons. Using this data, I built a neural decoding pipeline using Python (scikit-learn) and MATLAB-based preprocessing, including spike binning, normalization, train-test splitting, and confusion-matrix evaluation. In the study, I evaluated RF and KNN against traditional decoding methods, including population vector and maximum likelihood estimation.

Impact:

• This research will help to enable brain-controlled interfaces like prosthetic limbs, wheelchairs, and assistive devices for people with paralysis, spinal cord injuries, or limb loss.

• In healthcare & neurorehabilitation, it will support stroke rehabilitation and motor recovery by decoding intended movements even when physical motion is impaired.

• In Human-Computer Interaction (HCI), it will advance hands-free, thought-driven interaction systems for augmented reality.

• It showed that classical ML models can outperform more complex approaches without large data requirements.

• Achieved up to 91.7% decoding accuracy with KNN, outperforming Random Forest (up to ~80%) on the same dataset.

• Contributed evidence to an underrepresented area of BCI research, supporting transparent, data-efficient modelling in neuroscience.

• Concluded that K-NN is particularly effective for decoding movement direction in moderate-sized neural datasets, likely due to its ability to capture local geometric structure in neural firing patterns.

• Found that Random Forest remains a strong alternative when noise and interpretability are priorities, despite its slightly lower accuracy.

Outcome:

• Submitted my research paper for peer review to Curieux Academic Journal and National High School Journal of Science (NHSJS). Currently awaiting editorial decisions.

• Authored a full-length, publication-style research paper documenting these findings.

• Developed a strong foundation in machine learning, data analysis, and scientific methodology.

• Gained experience with academic research, writing, and the peer-review process, working at a level typically seen in early undergraduate research

• This experience confirmed my interest in pursuing computer science and engineering at the intersection of AI, research, and human-centered applications, and it motivated me to continue exploring how technology can translate scientific insight into real-world impacts.