ANALYTICS / PYTHON
Retail Demand Intelligence
A chronological forecast comparison connected to an inventory decision, using synthetic data.
← All projectsObjective
Compare a simple demand baseline with a model and examine what the difference means for replenishment.
Tools & architecture
Python, pandas, scikit-learn, HTML dashboard. Synthetic observations → chronological training/holdout → saved forecasts and metrics → dashboard.
Development & challenges
The synthetic dataset contains 4,992 weekly observations. Training uses weeks 5–84; weeks 85–104 form a rolling one-week holdout.
Security considerations
Use synthetic data and avoid claims of access to retailer records. Keep evaluation timing explicit.
Testing & verification
The saved holdout results show 26.8% wMAPE for the baseline and 9.8% for gradient boosting. The October 21, 2026 record contains five checks.
Outcome & limitations
A synthetic experiment, not measured store savings. Contemporaneous weather availability is a stated evaluation limitation.
What I learned
A better model score does not by itself establish a better stocking decision.
Source code is maintained in my private GitHub repositories.