DISCUSSION

• Scenic beauty estimation problem

  
  

  • Both C4.5 and Bayesian decision tree systems perform better than PCA system on training data but worse on test data

    
    

  • C4.5 produces 0% error rates on all closed loop tests but performs worse on open loop tests, which suggests the system might be memorizing the training data

    
    

  • C4.5 seems to produce more errors when classifying data with continuous attributes

    
    

• Surname pronunciation problem

  
  

  • C4.5 does not perform as well as the Bayesian approaches on small data sets

    
    

  • The results produced by the C4.5 trees are comparable to the Bayesian trees

    
    

  • Further investigation on different pruning and smoothing methods might prevent the decision trees from overfitting data