Last year, on college, I started working on a bot to consistently win money in daily fantasy basketball. I think I went negative before I paused working on it.
Pulled data from nba stats API using a Python library, created features in Python, then fed the data into pretty basic machine learning models. Only achieved slight gains over just assuming a player's performance was going to be his average over the past year.
But a big part that was pretty effective and not even predictive related was using a modified knapsack algorithm to fit the best predicted value into one day's roster.
