Parkinson's Disease Treatment Research

What this was

My MSc dissertation at Northumbria University, run with a clinical research group studying whether cues delivered by smartphone or wearable help reduce drooling in people with Parkinson’s disease. Data came from 70 participants across several groups and experiment phases.

To be clear about what this is: academic research, not a client engagement or a deployed product. No publication resulted.

The approach

The wearable data arrived as raw binary from the devices, so the first job was reading the hardware documentation and building an extraction pipeline. That work surfaced the project’s biggest problem — the device software had corrupted a chunk of the collection, and critical data was missing. Collection restarted, but with weeks left before submission I pivoted the analysis to the phone-collected data only.

From there: usage-pattern analytics across time of day, day of week and age group; four hypotheses tested formally, including phone-versus-wearable effectiveness; and a Random Forest model predicting personalised cue timing. Everything ran locally in Python and scikit-learn.

What came of it

The model reached 65% F1. On my supervisor’s advice it was framed as a recommendation tool rather than a prediction system — an honest read of what that score supports. I defended the work to my supervisor, the clinicians on the project, and a data science professor, and the supervisor used the results to argue for further investment in the wearable side.

The dissertation was graded 90/100, and the experience — clinical research methodology, working with healthcare professionals, and getting from raw hardware output to defensible analysis under time pressure — is what pointed the rest of my work toward healthcare.