# AnalyticBridge

A Data Science Central Community

# Logistic regression

Trying to compare 2 methods for churn prediction . 1st method is based on churn attrition calculation by extrapolating churn attrition rate and 2nd method is based on churn prediction by logistic regression. Obvious difference is that the fiurst method gives the headcount but the second method can give individual prediction. Another is that 2nd method is more realistic as it is based on number of realistic variables. Interested to know some more comparisons.

Views: 136

Comment

Join AnalyticBridge

Comment by Ralph Winters on November 16, 2010 at 8:36am
I would suggest performing some walk-forward testing with a validation sample to see which is the better performing modeling method.

-Ralph Winters
Comment by Minethedata on November 16, 2010 at 1:04am
Ralph,
Actually i learned that the group who is doing this has wrongly called this extrapolation but actually what they are doing is calculating the attrition rate per month from the data collected till date(time series) and multiplying it by the number of months to get the number predicted to churn at the end of these months. OR they just take the current attrition rate and find out if this exists what will be the attrition at the end of the year.
Comment by Ralph Winters on November 15, 2010 at 8:19am
How are you doing the first method? Extrapolations sounds like you have a time series on churn attrition for different groups and you are extending the line into the future.

-Ralph Winters
Comment by Minethedata on November 15, 2010 at 2:36am
Tom, Ralph please give your inputs :- Trying to compare 2 methods for churn prediction . 1st method is based on churn attrition calculation by extrapolating churn attrition rate and 2nd method is based on churn prediction by logistic regression. Obvious difference is that the fiurst method gives the headcount but the second method can give individual prediction. Another is that 2nd method is more realistic as it is based on number of realistic variables. Interested to know some more comparisons.