Adaface Sample Data Science Questions

Amazon's electronics store division has over the last few months focused on getting customer feedback on their products, and marking them as safe/ unsafe. Their data science team has used decision trees for this. 

The training set has these features: product ID, data, summary of feedback, detailed feedback and a binary safe/unsafe tag. During training, the data science team dropped any feedback records with missing features. The test set has a few records with missing "detailed feedback" field. What would you recommend?

A: Remove the test samples with missing detailed feedback text fields
            B: Generate synthetic data to fill in missing fields
            C: Use an algorithm that handles missing data better than decision trees
            D: Fill in the missing detailed feedback text field with the summary of feedback field.

Your friend T-Rex is working on a logistic regression model for a bank, for a fraud detection usecase. The accuracy of the model is 98%. T-Rex's manager's concern is that 85% of fraud cases are not being recognized by the model. Which of the following will surely help the model recognize more than 15% of fraud cases?

Your data science intern Rox was asked to create a decision tree classifier with 12 input variables. The tree used 7 of the 12 variables, and was 5 levels deep. Few nodes of the tree contain 3 data points. The area under the curve (AUC) is 0.86. As Rox's mentor, what is your interpretation?

A. The AUC is high, and the small nodes are all very pure- the model looks accurate.
            B. The tree might be overfitting- try fitting shallower trees and using an ensemble method.
            C. The AUC is high, so overall the model is accurate. It might not be well-calibrated, because the small nodes will give poor estimates of probability.
            D. The tree did not split on all the input variables. We need a larger data set to get a more accurate model.