- There is no such value which makes us sure about the scores are biased.
- Ideally, we can get the value by plotting the values of evaluation metrics for different parameter values of the model.
- For example, let us take a decision tree with varying depth of the tree.
- For each depth value(say 2,3,4,5,6,7,8...), we plot both training and test error.
- We could get a plot something like the above and when the train error and the test error start to deviate, we say the model has started to over-fit.
- So, we choose the value where the vertical line is shown in the plot.
Comments
0 comments
Article is closed for comments.