Project case study · AI / data science
AgroValue
An agricultural commodity price forecasting project built around historical commodity data, government mandi data, time-series analysis, model comparison, and a Streamlit dashboard.
Data workflow
The repository describes a workflow from Government of India AGMARKNET daily mandi data through collection and validation, commodity/market time series, feature engineering, forecasting, backtesting, and dashboard presentation.
Fields in the daily archive include state, district, market, commodity, variety, grade, arrival date, minimum price, maximum price, and modal price.
Forecasting and evaluation
The current app README describes a monthly dataset and a comparison between Seasonal Naive and SARIMAX. Evaluation uses a chronological holdout and reports MAE, RMSE, and sMAPE.
Chronological evaluation matters for time series because random splitting can leak future information into training. Results should be interpreted alongside the chosen holdout period and baseline.
Reproducibility
The repository documents installation through requirements.txt, running the dashboard with streamlit run app.py, and collecting daily data with python dataset/fetch_daily.py. Daily collection requires a data.gov.in API key configured as a secret or environment variable; credentials should not be committed.