Machine Learning to aid my Fantasy League Picks

I'm trying to apply machine learning to the data in my sports fantasy league, to see if I can't build a model to make my picks for me.

The rules are pretty simple:

• You make a pick on the winning margin of a real life sports event.
• If you are within 5 points of this margin, you get 0.5 points.
• If you are the closest person in your pool to the actual winning margin you get 1 point.

A pool is just your group of friends, and is normally between 5-25 people. Pool sizes are static for the whole season (18 weeks and about 18x6 games/picks). You can't see other people's picks until you have made your pick for that respective game. You can make the same pick as somebody else.

There are two exclusive strategies.

• You go really high or really low with your pick and hope you are on the edge of the group of picks. That way if the game ends with a high or low score you get a full point. For example if everyone chooses between 3-12 for the winning margin and you pick 20. If the game ends with a score of 50 then you get 1 point.
• You try and hit the actual winning margin of the game dead on, thereby increasing your chances of getting a margin point of 0.5 (easier to get, but less points).

GRAPH: This graph shows picks made by a large number of people for a single game. The horizontal axis is the margin, so to the left of 0 is Team A winning, and to the right of 0 is Team B winning. The vertical axis is the amount of people that made that specific pick.

Although the data has been aggregated, the people are all actually playing in small pools of 5-25 people. You can see based on that graph if I wanted to try and be the closest to a winning margin a good pick would be 17. That would put me on the high side with a low probability of having anyone higher.

What is surprising is how people have a tendency to pick certain numbers. I keep seeing people pick the number 4, 6, 8, 11, 13, and 16 way more than other numbers.

So here is the problem that I want to solve. Given that the bookies say Team A will beat Team B by a margin of X, and given that I have Y other players in my pool, what should my pick be?

Will I be able to solve this with machine learning or statistics if I have enough historic data?