分类模型的评价 - Lift Gain ROC AUC Lorenz_etc.pdfVIP

KXEN ANALYTIC FRAMEWORK Training - Charts The Data Mining Automation Company TM KAF VERSION 4 AGENDA 1. PRINCIPLES 2. DETECTED 3. LIFT 4. CUSTOMIZED 5. STANDARDIZED 6. ROC 7. LORENZ ‘GOOD’ 8. LORENZ ‘BAD’ 9. DENSITY ‘GOOD’ and DENSITY ‘BAD’ 10. INDICATORS 2 AGENDA 1. PRINCIPLES 2. DETECTED 3. LIFT 4. CUSTOMIZED 5. STANDARDIZED 6. ROC 7. LORENZ ‘GOOD’ 8. LORENZ ‘BAD’ 9. DENSITY ‘GOOD’ and DENSITY ‘BAD’ 10. INDICATORS 3 Principles Churner 30 Customers and 10 Churners Non Churner Churning Rate = 10/30 = 33% Score Max Score Min The Population is ordered by Decreasing Scores Values 4 Principles : Wizard Model Score My Model Score Max Min Score Max Score Min Customers with the highest Scores are Churners 5 Principle : Random Model Score My Model Score Max