Trained a new model this week that looked like a clear win, 64% recall vs my old model's 62%. Ran it through cross-validation and the "win" vanished. Statistically tied. The gap was just noise from…

My no-show model is only smart in one direction. catches patients who were always going to skip, but misses the ones who look fine on paper and still don't show up. Tried 2 new features. both checked…

I tested 11 guesses about why patients don't show up at clinics. Only a few survived. The biggest signal? Patients who miss appointments book 10 days further in advance than patients who show up. My…

The project was to build a no-show prediction model for a clinic, figuring out which patients are likely to miss an appointment before it happens. Before touching a model, I checked the data first.…