In my experiment, I used Decision Trees to predict whether participants will pass or fail, and I will provide feedback to them based on the reason for their failure. The Decision Tree includes three attributes. My question is: If a participant is predicted to fail, how can I determine the reason for their failure? ng? Thanks for your attention.
Is the node just before the failure the reason for the failure? What is the best criterion to choose among the three attributes associated with failure?
For your first question, I'd recommend reading 3.2 of this document: https://people.eecs.berkeley.edu/~brewer/papers/icac2004_chen_diagnosis.pdf
For the second question, I'd say that the node just before the failure isn't necessarily the reason for failure itself. It's the path leading to this node that specifies the reasons. The final leaf node simply provides the prediction based on the accumulated conditions from the root to that point.
Lastly, for the third question it depends on your goals and the provided data. I'd recommend trying both Information Gain and Gini Impurity. After that you can compare both results and make a decision. Also, you can look at the tree and evaluate the importance of each attribute
10.1109/ICAC.2004.1301345 . In case the link ever goes down (it's a Berkeley link so I doubt it ever will
