Motivating Question

Where are we?

Within the supervised learning framework, we have a categorical response variable \(y\) and a set of potential predictors \(x\). For example:

• y = vote / don’t vote, x = (age, party id, …)
  
• y = spam / not spam, x = (# of $, # of !, …)
  
• y = human / car / plant, x = (speed, shape, …)

We have the following goals:

• Build a classification model
  We’ll use the following techniques to build classification models of \(y\) from predictors \(x\):
  • parametric techniques
    • logistic regression (with or without LASSO!)
      
    • support vector machines (optional)
  • nonparametric techniques
    • K Nearest Neighbors (KNN)
    • classification trees
      
    • random forests and bagging

• Evaluate the quality of a classification model
  We’ll use the following metrics and tools to evaluate the quality of a classification model:
  • overall accuracy, sensitivity, & specificity
    We can approximate these metrics using in-sample and cross validation techniques.
    
  • ROC (receiver operating characteristic) curves