Case study · Health research, Universidad Icesi · 2023 to 2025 · Unpublished

Pesticides and Parkinson's disease: geospatial machine learning

At Universidad Icesi's I2T lab I worked on a question doctors in the region kept raising: do Parkinson's patients who live close to crops, and so to pesticides, have worse symptoms? I georeferenced patients from field surveys and built the spatial and machine-learning analysis.

Context: I2T lab, Universidad Icesi
Data: 6 field brigades · ~100 surveys
Status: Unpublished research

From field surveys to coordinates

The data came from six field brigades and about 100 patient surveys. I geocoded each patient's home and measured its distance to farmland and to crop mosaics, using land-cover layers in QGIS.

Method

Nearest-neighbour clustering grouped patients into three groups by proximity to crops. I then trained an XGBoost classifier to predict disease stage (Hoehn & Yahr) from geographic and environmental features, and used feature importance to see which ones mattered.

Results: unpublished

The research is unpublished, so detailed figures stay with the I2T lab. In summary, proximity to crops had a statistically significant association with the disease measures, but it did not change the gradient of Hoehn & Yahr stage or of the other Parkinson's measures. I presented the work to the lab's doctors, engineers and designers.