Yes, but that’s not much of a feature. You don’t have to detect anything, you can just use a different prediction model for each context, and train each separately.