I'm doing time series forecasting for the sales of an assortment of items. The data set I have contains two years of sales record of items that are may or may not still be selling. The challenging part is that the product life cycles are usually one year, after which the sales would stop and the new items would be introduced. So anything with more than a year of sales history is likely to be outside of my forecasting scope as it would soon be removed from the assortment.
Intuitively, I'd think the algorithm should fit a model on all the active items (those with sales numbers) within a certain timeframe and extrapolate the components for the assortment as a whole, and then predict the sales of the currently active items using their historical data.
I'm new to ML and have tried several time-series forecasting methods in Python but so far none of them seems to yield any meaningful results. I'd appreciate it if you can point me in the right direction.