PROJECT / MULTIVARIATE-LAB
From pattern to model.
What can we discover by looking at several variables together?
An app for exploring their relationships, finding similar profiles and trying prediction models.
What you can explore
Choose from three examples: metabolic profiles, student performance and blood tests. Select the variables you want to explore, see how they relate and find records that look alike. PCA charts bring many variables into a simpler view, while clustering lets you explore different profiles. In the student and blood-test cases, you can also choose a prediction task and compare models. I built this R and Shiny app from my multivariate analysis coursework at the UOC.
What the project tells us
The aim is to move from spotting a pattern to checking whether it helps make predictions. You will see prediction errors and comparisons with a simple reference, measured on data the model did not learn from. The public demo uses synthetic data, so you can experiment without exposing the original academic files. The results below come from the original coursework and may differ from the demo. The health examples are educational, not diagnostic tools.
EXECUTED EVALUATION / 2026-10-01
Results use an academic copy without removing observations based on residuals. Improvement classification uses the previous score available before the exam; it does not show that an intervention causes improvement. The coronary test has only three positives and does not validate a clinical tool.