Chapter 6 Engineering Numeric Predictors

Learning objectives:

• Learn about common issues and techniques when handling continuous predictors
  
  
• Often dealing with continuous predictors can be corrected by the model you select

  • Skewed data? Use tree-based methods

    • K-nearest neighbor and support vector machines should be avoided

  • Highly correlated variables? Use Partial Least Squares

    • Multiple linear regression and neural networks should be avoided
      
      
      
• Feature Engineering techniques to:

  • Address problematic characteristics of individual predictors

  • Expand individual predictors to better represent complex relationships

  • Consolidate redundant information