Week 2 Application Assignment of Predictive Modeling and Analytics

Consider linear regression models with ExpirationMonth column as the target variable. Find the model with one predictor variable and the highest R-squared. Consider the following set of predictor variables: Salerank, X2013USSales, X2013WorldSales, NumStores,RewardSize, and ProfitMargin. Which variable did you choose?

What is the estimated intercept coefficient of the model?

What is the estimated slope coefficient of the model?

Data transformation is a great way to improve model fit. Now consider the log transformation for the model identified in the previous question. [Hint: Use log function to create the transformed columns.] You can choose to transform neither of them, one of them, or both of them. You should have four different models.

• Model 1: neither variable is transformed; this gives you the same model as in the previous question.
• Model 2: only the target variable is transformed
• Model 3: only the explanatory variable is transformed
• Model 4: both variables are transformed.

Report the R-squared values of all four models.

What is the R-squared for Model 1?

R-squared for Model 2 is ( report answer using 4 decimal places i.e.  ):

R-squared for Model 3 is ( report answer using 4 decimal places i.e.  ):

R-squared for Model 4 is ( report answer using 4 decimal places i.e. ):

Which model gives the best fit based on the R-squared value?

Our analysis so far shows that variable transformation does not improve the model fit. Another way to improve model fit is to add more explanatory variables on the right side. Again consider the following set of predictor variables: Salerank, X2013USSales, X2013WorldSales, NumStores, RewardSize, and ProfitMargin. Add one more variable to the best model you identified in the previous question. Which variable will you add? Hint: The correct additional variable gives the highest R-squared value.

What is the R-squared for the model with the additional variable added ( report answer using 4 decimal places i.e.  )?

One way to figure out whether a linear regression model explains a particular data point well is to look at the residual. For which retailer do you have the highest absolute value of residual based on your result in the previous question?