medication-stock-forecast

End-to-end pipeline for forecasting medication stock needs at healthcare facilities using time-series AI.

Initially focused on hypertension; easily extendable to diabetes and other therapies.

Repository Structure

medication-stock-forecast/
├── data/
│   └── metabase_report.csv         # Raw Metabase export (git-ignored)
├── scripts/
│   ├── prepare_prophet_input.py    # Transforms raw CSV → prophet_input.csv
│   └── run_forecasts.py            # Fits Prophet and writes forecast CSVs
├── outputs/                        
│   ├── prophet_input.csv           # Model-ready data (git-ignored)
│   ├── forecast_results_12mo.csv   # Raw forecasts (git-ignored)
│   └── forecast_results_pivot.csv  # Pivoted forecasts (git-ignored)
├── .gitignore                      
├── requirements.txt                # Python dependencies
└── README.md                       # Project overview and instructions

Setup

1. Clone the repository

   git clone <repo-url>
   cd medication-stock-forecast

2. Create and activate a virtual environment

   python3 -m venv .venv
   source .venv/bin/activate

3. Install dependencies

   pip install --upgrade pip
   pip install -r requirements.txt

Usage

1. Export data from Metabase

   • Open this question in Metabase:
     Monthly Hypertension Patient-Days by Drug Class for top 5 uhc facilities
   • Click Export → CSV → save as data/metabase_report.csv

2. Prepare data for Prophet

   python3 scripts/prepare_prophet_input.py

   • Reads data/metabase_report.csv
   • Writes outputs/prophet_input.csv.

3. Run forecasts

   python3 scripts/run_forecasts.py

   • Reads outputs/prophet_input.csv
   • Writes outputs/forecast_results_12mo.csv (long)
   • Writes outputs/forecast_results_pivot.csv (wide)

4. Inspect outputs

   • outputs/forecast_results_pivot.csv has one row per month & facility, columns for each drug class forecast.

Extending

• To support diabetes or other conditions, update the Metabase query and adjust the class_cols pattern in prepare_prophet_input.py.