Machine Learning Applications in Fantasy Basketball

This paper is an attempt to apply machine learning to fantasy sports in order to gain an edge over the average player. Basketball players’ fantasy scores were predicted using a linear regression algorithm and stochastic gradient descent as well as a naive bayes classifier with discretized state space. A team of eight players was then selected by framing the problem as a constraint satisfaction problem. Regression optimizations were complex, but an advantage of around 8 percent was gained over regular users of DraftKings, meaning with a large enough volume of teams, the algorithm will make money. Future investigation is necessary into factoring in riskiness of players, as well as purposely pursuing players who are less frequently selected in DraftKings contests.