In this exercise, you will recreate some of the analysis done in this chapter for the COMPAS recidivism case, using the data in COMPAS-clean.csv. After partitioning the data into training and validation sets, fit a logistic regression model, a single decision tree, and a random forest model.

a. Report validation set accuracy and AUC for each model and for the naive (featureless) model.

b. Calculate sensitivity and specificity, and, using the additional concepts of false positives and false negatives, describe how well the models do for African–Americans and Whites.

c. Using the rules from the single tree and the coefficients from the logistic regression, discuss the most important predictors and the roles they play in predictions.

d. Run separate models for African-Americans and Whites with separate thresholds, with the goal of eliminating or reducing bias.