PESSIMISTIC PRUNING

• Use statistical methods to calculate the error rate associated with each node and adjust the tree to reflect bias

  
  

• Pessimistically increase errors observed at each node using statistical measurements to encourage pruning

  
  

• Advantages over other pruning methods

  
  

  • Builds only one tree

    
    

  • Does not require held out training data for error estimation

    
    

  • Provides a more reliable tree when data is scarce