Project · forecasting
Can a learned forecaster beat a mechanistic baseline at predicting disease progression — and where does it break?
Sequence models for multi-step forecasting, evaluated against persistence, statistical, and compartmental baselines under rolling-origin validation.
Question
[Forecast target and horizon; what "beat" means quantitatively.]
Baselines & validation
[Baseline table — persistence / ARIMA / compartmental vs. LSTM. Rolling-origin scheme described. This is the credibility, state it plainly.]
Where it breaks
[Regimes the model fails in — turning points, sparse regions. Naming failure modes is the investigator signal.]
Artifacts
[Forecast-vs-actual plot with error bands · repo link.]