Non monotonic feature selection decision
Why is it bad to have a non monotonic feature when attempting to determine an output. For example why would it be bad to select a temperature for determining a patients health?
In opposit I say.
Suppose we consider two features F1(time),F2(temprature) that F2 varies monotically with F1.
- Entropy of F2 given F1 is low.
- Correlation of F1 and F2 is too high.
Don't worry cause the first feature selection algorithm removes it! :)