I think it's important to remember that the 538 projection is the average of simulated outcomes.
Perhaps the 538 model looks at the Celts 48 wins last year and thinks that, all things considered, that was closer to a best case outcome for that team. Perhaps it also thinks that the loss of Turner and Sullinger is significant enough to substantially eat into any potential improvement attributable to the addition of Horford, Brown, and Green.
The range of significantly likely outcomes for the Celts in the eyes of their model could be, say, 44 to 55 wins, with the median skewing toward the lower end of that range.
Personally I think that 48 wins was a very good outcome for last year's team. At the same time, I think they're now head and shoulders above all the other Eastern Conference teams outside of Toronto and Cleveland, so winning less than 50 games would be an underachievement as far as I'm concerned.