usually minitab handles missing data. just interpolate it thats how machine learning works mostly, instead of predict, fill. for example make a graph between data that creates bridges between data, then for extension the graph should give an aproximate value around the missing data.

I would add to Ivan's already sufficient answer that in order to do the (linear?) interpolation, you have to be sure that those 20 influencing points follow the same "law", i.e. that interpolation as such is indeed possible.

Other than the concept of "interpolation" and "approximation", you can google also "imputation" - based on a probabilistic distribution, you fill the missing values by the most likely values. You even might get an associated probability for each of the 10 missing points, corresponding to the confidence you can put into it