Branch8 consulting work

Actionable insights and retention prediction modeling from e-commerce seller profiles

Transcript

1. Sandra Blevins Insight Data Science Insights in Action: Building e-commerce seller profiles to predict retention
2. ? eBay Amazon
3. Goal Keep sellers active and happy! Build seller profiles, predict retention/churn
4. Export MongoDB Clean Python regex, NumPy, pandas, JSON Database schema Seller profiles Predictive modeling Python scikit-learn
5. Seller profiles Active Churn ? 7 days
6. Seller profiles Active 7 days Churn ?
7. Seller profiles Active 7 days Churn ?
8. Seller profiles Active 7 days Churn ?
9. Seller profiles Active 7 days Churn ?
10. Number of sellers Active sellers in Shanghai, Hong Kong, and Singapore
11. 300+ Number of orders Active sellers have more orders
12. 300+ Number of orders Active sellers have more orders Active!
13. Active sellers Inactive sellers
14. Sellers of electronics with active Malaysian customer base Deliverable: actionable insights on seller profiles
15. sklearn.ensemble.RandomForestClassifier 10-fold cross validation Accuracy = 93%, AUC = 71% n-orders is best predictor of activity n orders n platforms mean price
16. Organize database Codes and flowsheets Customer profiles New feature recommendations for modeling
17. Sandra Blevins Education: Physics Ph.D. 2015, The Catholic University of America, Washington, DC Dissertation: The Distribution of Volatiles in Protoplanetary Disks Love: travel, nature, art, yoga