Artes Marciales Mixtas en Español is thrilled to welcome Jason’s Picks back after a very long hiatus. For the sake of sustainability, Jason’s Picks will focus on fewer fights per event than before.
True to its experimental philosophy, Jason’s Picks has taken a different approach to this coming Saturday’s UFC on FX 7 predictions. This time it has introduced new variables that may or may not inject bias into the model – something worth testing, especially on an evidence based “study” like this. Here is how it works:
This blog’s Editor was asked to rate fighters on a scale from 1 to 10 – 1 being extremely poor, and 10 being extremely good – on three different categories: submissions, striking, and wrestling. In order to minimize the potential for bias, the Editor based the ratings mostly on fighters’ statistics, even though that was not necessary, since the model already contained the quantitative variables. Nevertheless, the Editor did exercise his discretion when rounding the ratings up or down, based on personal opinion about each fighter. Besides, using fighters’ statistics made it easier to come up with ratings.
One challenge with the Editor’s approach to the ratings, however, is that fighters’ statistics are just numbers that cannot take into consideration issues such as quality of competition and, perhaps even more relevantly, stylistic matchups: for example, a fighter may have 100% takedown defense, but it may be due to opponents’ poor wrestling skills or lack of interest in taking a fight to the ground. That is where the Editor’s subjective opinion becomes useful.
One of the objectives behind introducing those ratings as variables into the predictive model was figuring out whether the ratings become more important than the metric variables themselves. At least for UFC on FX 7 on January 19th, 2013, the striking ratings became important as a variable in the model, and the prediction is an 85% likelihood of victory for Vitor Belfort.
Generally, the prediction is due to Belfort’s good striking rating, which exceeds a certain threshold, as well as his submission attempt average, which is also greater than a certain threshold. It is important to keep in mind that the model is using only a small data set limited to middleweight fights, but that may be a better approach than the model’s previous attempts at predicting outcomes using machine learning algorithms.
Jason’s Picks can be tracked on http://jaysha101.wix.com/mmapredictiveanalytics
Editor’s Note: This is the seventh in a series of fight predictions by guest columnist Jason Noriega, an analyst at a web-based trading site. Jason participates in predictive modeling competitions, where he has placed in the top 10% using various machine learning algorithms. He also lectures around the United States.
The predictive model developed by Jason Noriega for Jason’s Picks is the copyrighted intellectual property of Jason Noriega, Nery Castillo-McIntyre, and Artes Marciales Mixtas en Español.