For the DataTalks.Club Machine Learning Zoomcamp, we are required to complete peer review of three other student's projects.
The level of learning in the course itself is amazing in general. In particular for a free course it is perhaps unmatched. Further, I am seeing that looking at other student's work is an additional and highly fruitful lesson.
The first peer review I did for the midterm project is yujiezhou001's heart-failure-prediction.
GitHub: yujiezhou001 - Heart Failure Prediction
#### The Goal
Use the results that are already collected to estimate each patient's risk before the angiography decision. The model answers one question: given these 11 measurements, what is the probability that this patient has heart disease? Here, "heart disease" means the angiography found at least one major artery narrowed by more than 50%.
#### How the model helps
The model learns from 918 past patients whose angiography results are known. It finds which combinations of measurements usually go with heart disease. For a new patient, it turns the 11 separate results into one consistent risk score that a clinic can act on
I find that this project was excellently done. It was brief, and to the point, and clear in every aspect. I ran train.py and notebook.ipynb in a GitHub Codespace, and everything checked out.