Parkinson's Disease Treatment Research

The Problem

Researchers at Northumbria University needed to evaluate the effectiveness of wearable and smartphone-based cues for treating drooling in Parkinson’s disease patients. Manual analysis of clinical trial data was time-consuming and unable to identify complex patterns across multiple treatment variables.

The Approach

Developed predictive ML models using Random Forests to analyse clinical trial data from wearable devices and smartphone applications. The models evaluated treatment schedule effectiveness by identifying patterns in patient response data across multiple cue types and timing configurations. The solution was deployed on AWS SageMaker for reproducible experiments, with comprehensive data visualisation for clinical interpretation.

The Results

The models achieved 65% F1-score in treatment schedule prediction, providing researchers with data-driven insights into optimal cue timing and delivery methods. The work contributed to ongoing medical research into non-invasive Parkinson’s symptom management.