PharmaForecast

Time Series Forecasting for Pharma

An ensemble forecasting system combining ARIMA, Prophet, and ML models for drug-demand prediction, shortage early warning, and adverse-event trend analysis, orchestrated by Airflow.

Methods & contributions

• Ensemble — series characteristics (trend, seasonality, intermittency) drive inverse-error weighting across ARIMA, Prophet, and ML members.
• Statistical rigor — ADF + KPSS stationarity logic, sMAPE backtesting, and Ljung–Box diagnostics.
• Feature engineering — lag, rolling, calendar, Fourier-seasonal, and trend features with no look-ahead.
• Operations — daily Airflow DAG with accuracy monitoring, drift detection, and retraining triggers.

Data flow

series → classify (trend · seasonality) → {ARIMA, Prophet, ML}
      → weighted ensemble → backtest (sMAPE) → monitor / retrain

Technologies