Do you mean not knowing the value of the feature in the future? In that case, you build the model on past values, and when you see the future values the model should use them then to make predictions.

Moving average is an alternative way to linear regression for time series data. I advise you to check time series predictions. There is also a Facebook library for that called prophet