Logistic Regression Doubt



Hi All,

I have a doubt on building logistic regression model. If I am building a response model whether a customer will respond or not to any particular email campaign. What will be the difference in my modeling technique if I am predicting he/she will respond in 3 months. Or if my model is predicting if he/she will respond in 6 months? what will be the difference in my modeling for these different cases?

Thanks in advance :slight_smile:


Hi @amrita4friends,

As far as I believe, the predictions of your model (3 months vs 6 months) will give the same result because time features are not inherently handled by the algorithms such as Logistic regression.

To explicitly use time features in your model, you have to use algorithms such as Hidden Markov model or Recurrent Neural Networks.

PS: If you have the time, please go through this PhD thesis which addresses this problem.


Just to complement @jalFaizy’s great response, even if logistic regression cannot deal with time series, time features can be manually added. If you have access to past data, you could add the mean of email responses in less than in 3 months and in less than 6 months. And it’s usually better to compute the mean by some sort of segment (gender, age, type of customer, etc.).

Note that time-specific models as the ones cited above will still work better, but this kind of hack is good enough to capture simple time patterns.