I know i am asking a very generic question but this is something that i encountered in one of my projects. I am working on churn prediction for a bank and one of the features that i was using average cash in bank in past 6 months. Now if i plot the cash data per customer it will be very static in nature (atleast for salaried people as you get salary once a month and transactions mostly happen on weekends). Hence i am seeing something like a step function as salary credited on last day of the month, then no transaction for a week which means cash stays the same and then a transaction on weekend hence cash goes down. Now this series will be non stationary as salary will be greater than spend (assumption). Also there could be random spikes for people with business (big payments). What is the best way to analyze such data? Please provide some hints atleast (or point me to some paper/article).
Thanks in advance.