Decision Tree Learning Systems

• C4.5
  
  • Quinlan, ``C4.5: Programs for Machine Learning'', Morgan Kaufmann, 1993
  • ftp://ftp.cs.usyd.edu.au/pub/ml
  • Datasets at ftp://ftp.ics.uci.edu/pub/machine-learning-databases
• C5.0
  
  • Quinlan
  • Commercial version of C4.5
• CART
  • Breiman, Friedman, Olshen, and Stone
  • Builds binary decision trees
  • Exhaustive search for best tree
  • Choose split s that maximizes
    
    $$\phi(s/t) \;=\; 2 P_L P_R \sum_{j=1}^{classes} \mid P(j/t_L) - P(j/t_R) \mid$$
    
    

  • Handles noise and missing data
  • Commercial