Stock-out forecast
ORS demand at PHC Bhairavpur
Forecast uses OPD fever load, diarrhoea seasonality, local rainfall proxy and censored-demand correction for prior stock-out days.
AI demand planning
Forecasts that survive sparse, dirty government data.
ArogyaGrid uses model choice by data quality: gradient boosting where history is usable, seasonal baselines for sparse centres and intermittent-demand methods for slow-moving medicines.
6 days cover
Medicine demand
Model: quantile GBM
Used when facility/item history exists. Output is p50/p75/p90 demand so reorder logic can be risk-aware.
Intermittent items
Model: Croston/TSB
Used for slow-moving antibiotics, reagents and emergency stock where zero-demand days dominate.
Footfall
Seasonal baseline + events
Day of week, market day, weather, mela/harvest calendars and disease season features drive OPD predictions.
Cold start
Matched-centre borrowing
New or digitally weak PHCs borrow priors from similar catchment size, geography, facility type and IPHS norm.