If I have to predict the weekly sales of the product from the past. My data is at weekly level for the product and has about 5 years of data i.e, 260 data points and have about 20 (independent) features too for each weekly sale.
my sales have yearly seasonality, seasonality also slightly based on some of the features.
I believe because of seasonlity, this is a problem to model as stateful lstm. But, instead of stateful data (to identify trends from the past year), can I use stateless LSTM ?
In case of stateless, I can use extra derived features like previous year sales, last month sales (lagged target) and features' values for previous month and previous year (lagged features) to account for the dependency on the previous year sales values.
Does this approach ( modeling stateful lstm task as stateless lstm with extra lagged features) sound reasonable and valid.
Also, is the data sufficient either for stateful or stateless lstm task. How much data do I need if I Want to use stateful or stateless lstm for this task.