|
2009-2010 NBA Regular Season Prediction Record |
HOW THE "SCORE PREDICTION" MODEL WORKS?
The predictive model uses regression analysis to determine coefficients for the factors that might have impact on game outcomes. As the model continuously improves itself by employing some optimization techniques, it only focuses on predicting the nearest final scores individually for road team and home team.
HOW TO EVALUATE THE RESULTS?
Since the goal is to output nearest scores for road team and home team, criterias such as margin of victory, accuracy of picking the winner team should not be primarily considered when evaluating the success. Though, the record for predicting winning teams has been tracked and is 337-162 (.675) since December 1, 2009.
An assigned grade for both road team and home team does a better job of classifying the results by taking points differential between predicted and actual score. Distribution of grades (team-wise and date-wise) has been documented at the prediction archive.
Here is a couple of charts that reveal success of the model by breaking results down to grades which indicate the accuracy of NBA score predictions:
HOW TO READ THE CHARTS?
Each score prediction is evaluated by a grade that measures the differential between predicted score and actual score. If this differential is in the "0-3" points range, it gets an "A"; for the "4-7" points range, it gets a "B" and so on.
|
|
Last Updated ( Sunday, 07 February 2010 )
|
