I was building a goal score prediction application for premier league(PL) games. While deciding for the data i dan into a dilemma. How much historical data should be used to train the algorithms? I have seen articles around internet where the users are taking as old as 2015 season data. I think its pretty irrelevant as the squads, managers, practically the whole team changes by then. I think we should consider last 2 seasons at most. Is this a correct approach? Also, if we end up using just 2 seasons of data in this case, does it mean that using Deep Learning on such problems is not worth while as the number of matches will be just 380x2?
1 post - 1 participant