I want to predict the number of people joining (inflow e.g. 4000, 5000, 6000 etc) online subscription. The dependent variable is ‘inflow in the first 4 weeks for a certain content title’ as this is what we would like to estimate for new future titles.

I’m taking several variables like time (year, month, day, etc), genre, content classification (age recommendations, etc.), number of episodes, language, etc. as independent variables. However, I have a number of variables which are binary (0, 1).

1- I am wondering what sort of statistical/machine learning method/tool I can use for this analysis?

2- How can I inference which variables contributes most to inflow?

Could someone please guide me through the analysis process? I would appreciate your time and useful insight.

Thanks in advance


So this is a very basic question. One could write a whole book to this answer as there are thousands of different methods. As a beginner, I can suggest you to read the book 'Introduction to Statistical Learning' which is freely available here: https://wwwbcf.usc.edu/~gareth/ISL/ISLR%20First%20Printing.pdf

If you are bored of reading there are also videos explaining each chapter. https://www.youtube.com/user/dataschool/playlists?sort=dd&view=50&shelf_id=4

Concepts that you MUST understand in my opinion (and that are covered in the videos) are:

  • Classification vs Regression (it seems that what you want is a regression)

  • Linear Regression and Logistic Regression (the latter one might be not applicable for your case but still its good to know what it is)

  • Resampling Methods such as Cross Validation

So if you watch the videos on these topics you will be ready for your task. The videos also cover examples in the R programming language so you can easily reproduce it with your data (make sure to get RStudio if you use R).

If you get this, you can continue with more advanced methods such as the Lasso regression, Decision trees and Random forests etcetera. All these methods are able to cope with binary predictor variables. Just check the performance for different methods and choose the best performing one. For all these methods there are importance measures of the variables. For this some R libraries have been created. Just use google to find those.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.