Please enable JavaScript.

Coggle requires JavaScript to display documents.

Ch. 5 Overfitting (Overfitting - tailoring the model to the training data,…

- - - - Splits data set into K partitions called folds.
        
        Folds are chosen to be used as test data. After finished we will have performance estimates on all folds and we can compute average and standard deviation
  - - - "sweet spot" on a fitting graph showing the tree with training data and holdout data. This is what we want, but there is currently not a procedure to find the "sweet spot"
        
        2 Techniques to avoid overfitting:
        
        stop growing the tree before it gets too complex
        
        specify a min number of instances that must be present in a leaf
        
        conduct a hypothesis test at every leaf to see if the information difference is due to chance. If so then the split is not accepted
        
        grow a complex tree and then prune it back
        
        Replace a set of leaves with one if it does not reduce accuracy