Reference no: EM133943145
Assignment: Political Persuasion, CART, Random Forests, and Uplift Modeling
Question
Overall Effect of the Flyer.
Overall, how well did the flyer do in moving voters in a Democratic direction? (Look at the target variable among those who got the flyer, compared to those who did not.)
Exploring Predictive Relationships.
Explore the data to learn more about the relationships between the predictor variables and MOVED AD using data visualization. Which of the predictors seem to have good predictive potential? Show supporting charts and/or tables.
Train-Test Split and Baseline Accuracy.
Now, split the data into a training set and a testing set using the partition variable that is in the dataset. What is the accuracy on the testing set of a simple baseline method that always predicts "did not move" for every voter? Get world-class assignment help from subject experts.
CART Model.
Build a CART model to predict MOVED AD, using all of the other vari- ables (excluding opposite, VOTER ID, SET NO, and Partition) as inde- pendent variables. You should use the training set to build the model and use the validation set approach as covered in the course material to select between a few different reasonable parameter values.
Model Evaluation and Comparison.
Plot the CART tree. Which variables were used in the tree? Which variables appear to be the most significant?
What is the accuracy of the CART model on the test set? Now build a random forest model to predict MOVED AD, using all of the other variables (excluding opposite, VOTER ID, SET NO, and Partition) as independent variables. Again, use the training set to build the model. Select reasonable parameter values (you do not need to use validation for the random forest model). Then compute the accuracy of the model on the test set. How does it compare to the CART model?
Based on predictive performance, choose one of these two models (CART or Random Forest) as your final model.
Propensity Scores Under Treatment and Control.
A "propensity" is defined here as the predicted probability that MOVED AD = 1 for a voter.
Using your chosen model from Question 5(c), estimate two propensities for each voter in the test set:
P(MOVED AD = 1 | Flyer = 1): For each voter in the test set, temporarily set Flyer = 1, keep all other predictors unchanged, and compute the predicted probability from your model.
P(MOVED AD = 1 | Flyer = 0): For the same voters, tem- porarily set Flyer = 0, keep all other predictors unchanged, and compute the predicted probability from your model.
Report the two predicted propensities (for Flyer = 1 and Flyer = 0) for the first three records in the test set.
Computing Individual Uplift.
For each voter, uplift is computed as:
Uplift = P (MOV ED AD = 1 | Flyer = 1)-P (MOV ED AD = 1 | Flyer = 0).
Using the two sets of predicted propensities from Question 6, compute the uplift for each voter in the test set. Report the uplift values for the first three records.
Targeting Under Resource Constraints.
If a campaign has the resources to mail the flyer only to 10% of the voters, what uplift cutoff should be used to select the top 10% of voters?