Typically, when assessing customer churn, static data is used, for example, recent or average customer spending data, customer characteristics, etc. This approach does not allow tracking the activity of one client in dynamics: decrease/increase in his expenses, changes in location, tariffs, etc.
Question: Is there a modeling approach that takes into account such changes? My dataset looks like this: enter image description here
Each client has a different number of lines showing the dynamics of their activity. How can you predict whether a client will leave next month given all his data (all the lines that refer to him)?
Your are trying to solve a multivariate time series classification problem. You may solve this problem by using classical machine learning algorithms:
However you can not directly use those with your data. These algorithms are designed for static data.
The straightforward way to transform your data, is to concatenate informations of one pearson into a single line: ID, AGE, SEX, TENURE_01.01.2020 ... #of_minutes_01.09.2020
The second way to transform your data is to aggregate your data by ID. The most difficult part will be knowing how to aggregate your data to try to describe the dynamic of your data well.
For exemple, try to use :
Now you have a dataset with each id appear only once. I hope it will give you ideas for your problem.