We finally worked out the math to predict game winners by implementing our own TREND-SEARCH® algorithm. TREND-SEARCH® selects the best subset of trends from a superset defined in our database and then finds the team that matches the majority of those key trends in a given matchup. The selected teams are projected to win the game ATS.

We are using a logistic regression classifier within the TREND-SEARCH® algorithm because of its speed and accuracy. Though the accuracy varies by season we note that the logistic regression classifier never performs below 50%. We've trained our statistical analysis against 10+ seasons of data and have found a maximum accuracy of 64%.