The question
Can you predict how a patient does during rehabilitation and treatment from data that already exists? And which factors matter most?
What the data showed
The models pointed to the factors that mattered most for the course and how well they predicted the outcome. Classical statistical models were held up against modern machine learning, so you could see what actually added value.
What it could be used for
A basis for assessing a course early, with honesty about how certain the prediction was, so it could support a clinical decision rather than replace it.
Tools
pandas and tidyverse for data wrangling and feature engineering. Logistic regression, random forests, and SVM, evaluated with cross-validation and ROC/AUC. Visualization of variable importance and partial dependence.
The same method predicts outcomes and points to the factors that matter most in any dataset — for example churn, demand or risk — always with honesty about how certain the prediction is.