I have a time series which is short i.e it has only 7 points (7 months) where the data is measured(Margin), it has many multi level categorical attributes and one numerical attribute called Total_Margin. Also, there are multiple measurements for every point per subscriber. I want to use machine learning to train a model on 6 months of data to try and predict the 7th month.

I have thought hard about this problem and came up with a technique to sort of deconstruct this time series into one row for every subscriber where all the data for every subscriber is flattened into one row. A simplified version of the flattening for just one subscriber can be seen here:

enter image description here

I have some questions regarding the problem I have, I apologise if they appear uneducated but I don't have a background in statistics.

  1. What would such a time series be called? (literature) examples forecasting such a time series with or without machine learning in literature?

  2. Is my flattening process a valid approach? if yes what would it be called, I think its loosely based on target based encoding but I'm not sure as i was unable to find something on it.

  3. I think I introduce collinearity among predictors with my method? would that be accurate.


Your Answer

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

Browse other questions tagged or ask your own question.