tag:blogger.com,1999:blog-5902620336509647050.post5595948998426173267..comments2023-10-31T07:06:33.886-04:00Comments on Net Prophet: Iterative Strength of Victory (ISOV)Scott Turnerhttp://www.blogger.com/profile/03393071448515738228noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-5902620336509647050.post-2840128677004280592011-06-08T18:28:20.565-04:002011-06-08T18:28:20.565-04:00The details can be found here (http://netprophetbl...The details can be found here (http://netprophetblog.blogspot.com/2011/04/testing-methodology-redux.html) but the general idea is that for each game we have a record which has the rating for each team as well as the margin of victory (MOV). I pull out about 80% of those games and train a linear regression using those examples. The linear regression is basically an equation that uses the ratings of two teams to predict the MOV, e.g., it will be something like:<br /><br /> 12*Home Team's IMOV - 10*Away Team's IMOV + 6 = Predicted MOV<br /><br />Then that equation is tested on the remaining 20% of the games -- the equation predicts the outcome and then it is checked against the actual outcome. The MOV Error is the square root of the squares of all those errors. RapidMiner handles all of that automatically.<br /><br />I've found that home/away tweaking doesn't usually improve performance because the linear regression equation (see example above) derives a constant term to add or subtract (see the +6 above), and that always turns out to be more accurate than anything I can do a priori. And more complex home/away models (see http://netprophetblog.blogspot.com/2011/04/prophet-dick-vitale.html) don't add accuracy, either -- a tentative conclusion is that home court advantage doesn't vary much from team to team.Scott Turnerhttps://www.blogger.com/profile/03393071448515738228noreply@blogger.comtag:blogger.com,1999:blog-5902620336509647050.post-84853973244851969242011-06-08T16:48:34.288-04:002011-06-08T16:48:34.288-04:00How are you calcing the MOV error? Do you have al...How are you calcing the MOV error? Do you have all of your systems predict MOV? If so, how were you determining an MOV prediction in the RPI-like systems, i.e. without using score information?<br /><br />And I assume one tweak you have in mind for this system is to include a home/away adjustment?<br /><br />And I just saw the link to my site on your roll. Thanks! I'll add a link for you on mine...probablepicks.comhttps://www.blogger.com/profile/08285913697327878399noreply@blogger.com