FreshRetailNet-50k - Food Retail Demand Prediction
Massive food retail data on 50,000 references with out-of-stock tracking, weather, promotions and more.
4.85 million hourly examples, structured tabular format (CSV/parquet)
CC BY 4.0
Description
FreshRetailNet-50k is a unique benchmark for predicting demand in the food retail sector. It contains 50,000 detailed 90-day time series, covering hourly sales of perishable goods in 898 stores across 18 major cities. The dataset also includes information on stockouts, promotions, weather, and holidays.
What is this dataset for?
- Improving demand forecasting models in the food retail sector
- Test stock-out and latent demand detection algorithms
- Train large-scale contextual time series models
Can it be enriched or improved?
Yes, this corpus can be enriched with logistical cost data, margins or external data (events, regional trends). It is also possible to annotate it more finely to refine the detection of hidden breakdowns or to add confidence scores to sales predictions.
🔎 In summary
🧠 Recommended for
- Supply chain data scientists
- Pricing and forecasting teams
- Time series researchers
🔧 Compatible tools
- PyTorch Forecasting
- GluonTS
- Prophet
- LightGBM
- XGBoost
💡 Tip
For best results, aggregate hourly data by fixed time slots (e.g. morning/afternoon) according to the products
Frequently Asked Questions
Does this dataset cover several types of products?
Yes, it contains more than 860 references of perishable products, classified by categories and sub-categories.
Can we identify the effects of promotions in the data?
Absolutely. The dataset includes a promotion indicator and a discount rate for each entry.
Can it be used for real-time models?
Yes, each line is time-stamped and can be used for simulations or continuous deployments for real-time models.
