Thread by @lambda_0206

WOE & IV
WOE (Weight of Evidence) determines the predictive power of an independent variable in relation to the dependent variable.

WOE = ln (%non events/%events)

The independent variables is replaced by WOE - It is monotonic and continuous

#DataScience #WOE #IV
1/n

Steps-
1 - For Continous Variables Make Bins of the Data
2 - Min 5% Data in each bin and non zero events/non-events in each bin
3 - Missings to be grouped separately
4 - Use the WOE formula
5 - Regroup similar value bins.
WOE is monotonic

2/n

Benefits-
1 - Treats Outliers
2 - Missing Values are binned separately
3 - Dummy Encoding is not required
4 - Strict Linear relation of the WOE Values

3/n

In case of Zero event/non events, Adjusted WOE can be used-

AdjustedWOE = ln (((Number of non-events in a group + 0.5) / Number of non-events)) / ((Number of events in a group + 0.5) / Number of events))

4/n

IV = (%non events - %events) * WOE

Information Values tells us the power of prediction of an independent variable. It helps to rank variables on the basis of their importance.

5/n

IV < 0.02 suggests to not use the variable
0.02 < IV < 0.1 Weak Predictive Power
0.1 < IV < 0.3 Medium Predictive Power
0.3 < IV < 0.5 Strong Predictive Power

IV > 0.5 is suspicious

6/6

#DataScience #WOE #IV @AndrewYNg @Medium @kaggle