avanishkasar/DemandFlow: DemandFlow v2.0.0 - Revision 1 (RINENG-D-26-22476)

DemandFlow v2.0.0 - Revision 1 Reproduction Package

This release accompanies the revised manuscript RINENG-D-26-22476: Search-Aware E-Commerce Demand Forecasting with Google Trends: A Leakage-Free Evaluation under Controlled Simulation

What changed from v1.0.1

• Leakage-free pipeline: all features now use data from week t-1 only
• Real Google Trends data: synthetic fallback series replaced with actual downloads
• 6 benchmark models: naive, seasonal naive, SARIMA, SARIMAX, XGBoost, Ridge
• 52-week rolling-origin evaluation with bootstrap confidence intervals
• Statistical significance: Diebold-Mariano tests for all model pairs
• Robustness: 100 replications x 5 coupling strengths + 110 sensitivity runs
• Leakage audit: quantifies how each leaked feature inflated original accuracy
• Decision modules: launch-window and pricing analysis with stated assumptions

Reproduce

pip install -r requirements.txt
python run_all.py            # full run (~2 hours)
python run_all.py --quick    # main experiment + figures only (~5 min)

Citation

If you use this code, please cite the associated paper and this Zenodo archive.