Exploration and Application of Dimensionality Reduction and Clustering Techniques to Diabetes Patient Health Records

Abstract

This research examines various data dimensionality reduction techniques and clustering methods. The goal was to apply these ideas to a test dataset and a healthcare dataset to see how they practically work and what conclusions we could draw from their application. Specifically, we hoped to identify similar clusters of diabetes patients and develop hypotheses of risk for adverse events for further research into sub-populations of diabetes patients. Upon further research and application, it became apparent that the data dimensionality reduction and clustering methods are sensitive to the parameter settings and must be fine-tuned carefully to be successful. Additionally, we saw several statistically significant differences in outcomes for the clusters identified with these data. We focused on coronary artery disease and kidney disease. Focusing on these clusters, we found a high proportion of patients taking medications for heart or kidney conditions Based on these findings, we were able to decide on future paths building upon this research that could lead to more actionable conclusions.