Progress checklist

[yelselmiao]

3-4 Introduction

• [ ] Project Overview (background, Objective, Importance)

• [ ] Propaganda, advertising target population, renter selection, fee

• [ ] Dataset Overview

• [ ] nrow(test/train), n(varibale), reponse, predictors, continuous, categorical(level) 6-8 EDA

• [x] Missing Value

• [ ] Inbalance Class (Predictor & Response)

• [ ] Transforming

• [x] Outlier Process

• [x] Skewness

• [ ] Correlation 1-2 methodology (Briefly Explain)

• [ ] Set Baseline Model

• [ ] Variable Selection

• [ ] Ultimate Model; Model for Comparision

• [ ] Cross Validation (K-fold) 1-2 Conclusion

[yelselmiao]

Based on HJ's comments:

• [x] 2. page 4: column 2 header should be "#missing(%)" what does complete-case mean for gender or age? Does this mean you keep cases with both gender and age non-missing?

• [x] For missing on "first-browser" can you make missing a separate category?

• [ ] 3. page 7: multiple lines for a user in the original sessions database converted to actiontriplet1, actiontriplet2, ... in the cleaned sessions database. If userid is a field, does this mean that there are at most 20 actiontriplets?

• [ ] 4. I think you will have to merge the categories with small counts.

• [ ] 5. pages 11, 12, 13, 14: more than half 0s for some countries? Sqrt transform for y-axis would be better?

• [x] 6. page 16: for these features, categories with small counts should be merged with other categories.

• [ ] 7. Unbalanced classes will cause difficulties.

After reducing the number of classes for the response variable, maybe naive Bayes can be used. Please look at default-naiveBayes.pdf at the course web site for density estimation for continuous variables after transforming away extreme skewness.