This seems to be ripe for attack through ML models. It should be easy enough to create models and test it out on previous years to see its accuracy. May be some of us can do ensembles of models with bunch of best predictions (Netflix style!). Any idea where to get data sets for previous games?
If you had any success doing that on a full on march madness bracket, you could probably make a lot more than a Billion dollars (over time, don't be greedy) - betting on the sports betting scene, which is a (great) deal larger than $1B/year.
The odds of winning are 1 in 4,294,967,296 at random guessing. If your system can do 100 times better, that reduces it to 1 in 42,949,672. Or an expected utility of a measly 23 dollars, so don't waste too much time on it.
