Revenue Forecast + Deal Optimizer

An end-to-end decision support project for merchant operations:

• Forecast daily revenue for the next 30 days
• Recommend deal dates and spend intensity under multiple scenarios
• Translate model output into execution-ready, business-facing guidance

Why This Project

Merchants often schedule promotions without demand-aware timing. This project addresses that gap by combining short-term forecasting with scenario-based deal recommendation.

Results Snapshot

• Revenue rolling backtest MAPE: 6.50%
• Revenue within +/-10% accuracy: 83.33%
• Recommendation candidates after model filters: 60 of 120
• Execution-ready recommendation dates (positive profit + ROI proxy): 14
• Dominant strategy in top candidates: light_push (1.25x)

My Contribution (Goyeun Yun)

• Designed and implemented the XGBoost revenue forecasting pipeline
• Built feature engineering framework (calendar, lag, rolling, ad metrics)
• Implemented expanding-window rolling backtest with recursive forecast logic
• Built scenario-based deal recommender with growth, efficiency, and balanced scoring
• Packaged project for reproducible GitHub delivery

Team

• Marcie (Kaixuan) Ma
• Jisu Um
• Goyeun Yun

Quick Start

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

python rebuild_xgboost_pipeline/build_features.py
python rebuild_xgboost_pipeline/backtest.py --target Revenue
python rebuild_xgboost_pipeline/forecast_30d.py --target Revenue
python rebuild_xgboost_pipeline/deal_recommendations.py --target Revenue

Data and Reproducibility

• Included raw input: data/raw/synthetic_ecommerce_data.csv
• Generated feature tables: data/processed/
• Generated outputs: outputs/backtests/, outputs/forecasts/, outputs/recommendations/

Generated outputs are excluded from version control by default to keep the repository lightweight and reproducible.

Project Structure

data/
	raw/
		synthetic_ecommerce_data.csv
	processed/
report/
	executive-summary.html
outputs/
	backtests/
	forecasts/
	recommendations/
rebuild_xgboost_pipeline/
	build_features.py
	backtest.py
	forecast_30d.py
	deal_recommendations.py
	deal_scheduling_recommender.py
	utils.py

Portfolio Notes

• This repository focuses on the final delivery scope: Revenue forecasting + deal scheduling.
• Orders and sessions forecasting were intentionally deferred due to timeline constraints.
• Current validation uses a synthetic proxy dataset; next step is re-validation on production data.