Extract features of uneven time series and use it in cross-sectional study

Question

I have a data like below

subj_1 = 10,20,15,30,60,70,90 (in resolution of years - 2011 to 2018)

subj_2 = 10,20,30,40 (in resolution of months - Jan 2011 to Apr 2011)

subj_3 = 10,15,20,30,45,55,60,70,90,56,79,21,90,80,45,87 (resolution of months - Apr 2014 to July 2015)

These time series indicate the profit made by each subject (by selling product).

For subj_1, it means that he had only sale each year from 2011 to 2018 whereas subj_2 had only 4 sales in the year 2011 (no other sales after).

I don't think it makes sense to copy previous values to make time series even because only when there is a sale, he makes profit or loss. So, when there is no sale, we cannot just assume it was profit (by copying previous values to rest of the observations).

Now, my only question is given below

a) I would like to extract some useful features that can represent/capture the trend (and other info) from these series. I don't try to forecast anything out of the above data but derive some useful measures from the above data which can then later be used as feature (for each subject) in a cross-sectional binary classification problem.

Basically, I expect my output to be something like below (where I include time based feature as input to cross-sectional binary classification model). You can see that each subj has only row (with time based columns as well). What are some of the useful time-based features that we can extract from the above series?

Any suggestions that you can help me with?

Answer 1

It sounds like you are trying to do time series feature extraction. A great package in Python for doing this is called Time Series Feature Extraction Library. This package "computes over 60 different features extracted across temporal, statistical and spectral domains." (See paper.)