open_acc was broken into multiple discrete bins as shown below: 
There are many small number and woe curve is flat. I looked at number of observations and group them to following bins:In this project, I build a scorecard for the insurance modelling. The data is taken from lending club. Here, I am using years 2007-2014 for data and 2015 as testing.
For the insurance data modelling, I binned the continous variables using Weight of Evidence methods. For example, the continous feature open_acc was broken into multiple discrete bins as shown below: There are many small number and woe curve is flat. I looked at number of observations and group them to following bins:
open_acc:<=3
open_acc:4_5
open_acc:6_13
open_acc:13_17
open_acc:>17Then with the clean binned data, I used logistic regression to estimate the probability of good and bad policies. - Build logistic regression and got the p-values. - Feature selection using p_values - Logistic Regression with p-values selected features. - Do the model evaluation on test data. - Plot Gini Curve. - Plot Kolmogorov-Smirnov curve. - Then I built the scorecard. 
