Azure ML Lab Midterm 1 of CalStateLA
Improve Performance of AUC better than this
In order to improve AUC I used the followoing:
- Used Clip Values in order to get rid of outliers, such as 99999 or 99 in some fields. Particularly used Percentile of 90%
- Used Edit Metadata and changed most columns to categorical type, as they had few unique values, which clearly means that those are categorical.
- Added column in Select Column Database “DG1”, which was initially omitted. Looked like a year of birth and overall that was very valuable field for a model.
- Changed Machine Learnign model to Two – Class Boosted Decision Tree, which performed better that Jungle
- Added Permutation Feature Importance and got rid of several values of Score 0.