For my project, I am trying to predict return ( when a product in ecommerce sale is returned) rate of products. For the same of simplicity, assume I have 3 static features (dont change in time) and return rate for each product. Products that are in later stage of their lifecycles seem to be returned much less than products that are newly launched. Thus cross-sectional analysis on newly launched and staple products would be misleading and like comparing oranges to apples. How can design a model where I can 'detrend?' or account for this inherent shape of return rates.
My current idea is to use dummy variables indicating in which stage of lifecycle the product is.