RandomForrest ML

[native oxide]

I'm happy to help. What specific questions do you have?

[heady hill]

I would start framing the problem, do you have current loan data? How familiar are you with measuring performance, assessing a model, tuning it. Are you creating this model from scratch?

[gentle flume]

Hi, thank you so much. I already did this:

• Cleaned the data (from LendingClub)
• Removed unnecessary columns (I now have 43 columns of the 151 left)

After this, I did the following, but I am not sure this is the way to do it.

• Divided the data in X and y (X = data - loan status : y = loan status)
• Splitted X and y with train_test_split
• Took 5 best features using the SelectKBest tool from sklearn
• Defined the RandomForest model and fitted this with the selected data
• Evaluated with predict_proba(X_test
• Created a roc curve

The reason I am not sure this is the way is that my ROC curve is too good I think, which probably means it is overfitted?

Thank you again!

[gentle flume]

I do have hat but I am very new to all of this. I have some sample code from lectures but other than that, everything is from scratch.

[heady hill]

What is the orange line?

[gentle flume]

the orange line should be the performance per threshold

[heady hill]

How come it says feature selection?

[gentle flume]

because this is a name i gave just because i used the features from SelectKBest 🙂

[heady hill]

That’s a good roc curve

Is this predicting X train or X test

[native oxide]

^ + 1, Is this curve the performance on the train dataset or the test dataset?