|Last Week||2 star or better||2-1-0||66.7%||+0.8|
|Last 3 Weeks||2 star or better||5-4-0||55.6%||+0.5|
|View ATS picks for upcoming games »|
|Since 2008||3 star or better||463-403-11||53.5%||+17.9|
|Last 3 Weeks||All Picks||89-67-2||57.1%||+13.9|
|Last 500 Picks||3 star or better||266-230-4||53.6%||+11.8|
|Last 100 Picks||3 star or better||53-45-2||54.1%||+3.2|
|View Over/Under picks for upcoming games »|
|This Season||3 star or better||40-86-0||31.7%||+14.1|
|View Money Line picks for upcoming games »|
|Last 3 Weeks||3 star or better||76-10-0||88.4%||--|
|Last Week||3 star or better||29-4-0||87.9%||--|
|Last 100 Picks||3 star or better||86-14-0||86.0%||--|
|Since 2008||3 star or better||2826-712-0||79.9%||--|
|Last 250 Picks||3 star or better||199-51-0||79.6%||--|
|View Game Winner picks for upcoming games »|
The Decision Tree model, based on proprietary TeamRankings.com technology, is what statisticians sometimes refer to as a "black box" model. Its predictions are based on highly detailed analysis of a large data set that incorporates hundreds of variables related to past matchups, such as team stats, betting lines, and final scores.
Our Decision Tree models are all hypothesis tested. Once a promising model is constructed based on historical test data, we apply it to a second, independent historical data set to cross-check its performance. In total, our basketball Decision Tree models leverage approximately 7 seasons worth of test data.
This model applies a machine-oriented learning approach that goes beyond regression methods commonly used for game prediction. Rather than attempting to determine the expected impact that a specific variable or stat will likely have on a game's outcome, it seeks to uncover complex and subtle relationships between multiple variables.
Our math model predictions are subject to change up until an hour before game time. These models are designed to process and respond to breaking information, such as recent game results and betting line movement, and new information often alters picks.
All model pick tracking is based on the information and betting odds we have in our system as of one hour before game time. At that point, we "freeze" our model picks for the game in question and report performance on those predictions.