Best Target Market of Diabetic Patients - Data Driven Recommendations
- 1. Predict Target Demographics for Diabetic Treatments Anh Do and Stephano Muro
- 2. Problems ➔ 30.35 million Americans with diabetes (2015) ➔ Fierce competition in the market ➔ Competitive advantages come from knowing the demographics to target
- 3. Agenda ➢ Sampling ➢ Understand the Data - Descriptive Analytics ➢ Predictive Models ➢ Assumptions and Limitations ➢ Conclusions and Recommendations
- 4. Sampling ❖ Original data from CDC - more than 229,000 rows ❖ Random sampling and cleaning - sample used has 6957 rows ❖ Attributes chosen: ➢ Age ➢ Sex ➢ Education Level ➢ Employment Status ➢ Income Categories ➢ BMI (raw score) ➢ BMI Categories ➢ Physical Activities Categories ➢ Leisure Physical Activities
- 5. Understanding the Data ● 871 respondents with diabetes, 6086 respondents without diabetes ● Among diabetics: 13% males, 12% females ● Omit “sex” from classification models
- 7. Understanding the Data Employment status of diabetics and the chosen sample
- 9. Understanding the Data Level of education attained by diabetics compared to the total sample
- 11. Understanding the Data Likelihood of diabetes and whether respondents engaged in leisure time physical activities Measures of lifestyle activeness of diabetic respondents
- 12. Predictive Model: Logistic Regression Continuous variables: ● age, ● income group, ● education level, ● body mass index (BMI) categories, ● activeness Classification variables: ● sex, ● whether the person engaged in leisure physical activities
- 13. Predictive Models: Logistic Regression Hypotheses ● H0: Binary Logistic Regression Model cannot predict whether a person had diabetes based on selected continuous and categorical variables. ● H1: Binary Logistic Regression Model can predict whether a person had diabetes based on selected continuous and categorical variables.
- 14. Predictive Models: Logistic Regression Classification variables
- 15. Predictive Models: Logistic Regression Continuous Variables
- 16. Predictive Models: Classification Label: “Ever Told You Had Diabetes.” “1” meant “Have diabetes” and “0” meant Does not have diabetes. Attributes: Age, Education Level, Income Group, Employment Status, BMI category, Leisure Physical Activities, Physical Activity Category.
- 17. Predictive Models: Classification K-NN (K=10)
- 18. Predictive Models: Conclusions ● K-NN best predictive results ○ K = 10 ○ Accuracy of 87% ● Failed Models ○ Decision Tree ○ Random Tree ○ Random Forest ○ Support Vector Machines
- 19. Assumptions and Limitations ● Assumption: treated Type 1 and Type 2 diabetes as similar ● Sampling Error ○ “Sample of a sample” ● Limitations on tools (SAS & Rapidminer)
- 20. Conclusions Pharmaceutical Companies focusing on diabetes treatments should segment market based on demographics
- 21. Questions?
- 22. Assumption on Type 1 and Type 2 Diabetes 29.1 million Americans with type 2 1.25 million Americans with type 1 Since the majority is suffering from type 2, our assumption still made business sense.
Editor's Notes
- #2: Anh
- #3: Anh
- #4: Anh
- #5: Anh
- #6: Stephano 14%
- #7: Stephano Need percentage
- #8: Stephano
- #9: Stephano
- #10: Stephano
- #11: Stephano
- #12: Stephano
- #13: Anh
- #14: Anh
- #15: Anh
- #16: Anh
- #17: Stephano
- #18: Stephano
- #19: Stephano
- #20: Anh
- #21: Pharmaceutical Companies focusing on diabetes treatments should segment market based on demographics since descriptive analytics found relationships between diabetes and many demographic characteristics, and predictive models successfully predict if a person has diabetes. Anh