Quit Predictor

Random Forest classification to predict who will fail at quitting cigarettes

Transcript

1. QUIT PREDICTOR PUIFAI SANTISAKULTARM, PH.D.
2. HYPOTHETICAL SCENARIO: LET’S QUIT SMOKING
3. CAN WE PREDICT WHO WILL RECURRING DEBTSSMOKING? REPEATEDLY QUIT DO NOT SMOKE
4. CAN WE PREDICT WHO WILL RECURRING DEBTSSMOKING? REPEATEDLY QUIT DO NOT SMOKE x1 QUIT ONCE
5. CAN WE PREDICT WHO WILL RECURRING DEBTSSMOKING? REPEATEDLY QUIT DO NOT SMOKE x1 QUIT ONCE REPEATEDLY QUIT
6. PREDICTING CONSUMERS WHO WILL RECURRING DEBTS AUC 0.68 EVERYONE REPEATEDLY QUIT x1
7. 7M Records ANALYTICS PIPELINE
8. 7M Records ANALYTICS PIPELINE ~20 features ▸ # cigarettes/day ▸ Income ▸ Cigarette brand ▸ Occupation ▸ Date of last smoke ▸ … } } } Numbers Text Dates
9. ANALYTICS PIPELINE BIG DATA DATA WRANGLING
10. ANALYTICS PIPELINE BIG DATA DATA WRANGLING VISUALIZATION
11. SMOTE FOR IMBALANCED DATA x1
12. SMOTE FOR IMBALANCED DATA Original x1
13. SMOTE FOR IMBALANCED DATA Original SMOTE x1
14. SMOTE FOR IMBALANCED DATA Original SMOTE RANDOM FOREST CLASSIFICATION x1
15. DEBT ONCE AND RECURRING DEBT ARE VERY SIMILAR Different Similar REPEATEDLY QUIT QUIT ONCE x1 x1
16. DEBT ONCE AND RECURRING DEBT ARE VERY SIMILAR DIFFERENT REPEATEDLY QUIT QUIT ONCE x1 x1
17. DEBT ONCE AND RECURRING DEBT ARE VERY SIMILAR VERY SIMILAR DIFFERENT REPEATEDLY QUIT QUIT ONCE x1 x1
18. GROUP CHARACTERISTICS OVERLAPS EXERCISE PER WEEK x1
19. GROUP CHARACTERISTICS OVERLAPS EXERCISE PER WEEK x1
20. GROUP CHARACTERISTICS OVERLAPS EXERCISE PER WEEK x1
21. IMPROVEMENT IN PREDICTION EVERYONE AUC 0.79 x1 DO NOT SMOKE QUIT ONCE REPEATEDLY QUIT
22. IMPROVEMENT IN PREDICTION EVERYONE AUC 0.79 16% x1 DO NOT SMOKE QUIT ONCE REPEATEDLY QUIT
23. DELIVERABLES ▸ Predictive machine learning model to classify consumer segments
24. DELIVERABLES ▸ Predictive machine learning model to classify consumer segments ▸ Detailed technical blog
25. DELIVERABLES ▸ Predictive machine learning model to classify consumer segments ▸ Detailed technical blog ▸ Jupyter Notebook of my analysis codes
26. PUIFAI SANTISAKULTARM Ph.D. Biomedical Engineering, Cornell University
27. BUSINESS DECISION TO MAKE… AUC PRECISION RECALL 0.22 0.89 0.98 0.65 0.40 0.05 0.79 0.74 0.60 DO NOT SMOKE x1 QUIT ONCE REPEATEDLY QUIT
28. BUSINESS DECISION TO MAKE… AUC PRECISION RECALL 0.22 0.89 0.98 0.65 0.40 0.05 0.79 0.74 0.60 DO NOT SMOKE x1 QUIT ONCE REPEATEDLY QUIT
29. BUSINESS DECISION TO MAKE… AUC PRECISION RECALL 0.22 0.89 0.98 0.65 0.40 0.05 0.79 0.74 0.60 False Positive True Positive DO NOT SMOKE x1 QUIT ONCE REPEATEDLY QUIT