As Pat Barry showed with his takedown and submission defense against Christian Morecraft, fighters change. And so it is with predictive modeling, an important part of which is learning from past mistakes, and using those mistakes in order to make improvements to the model continuously. To that end, I am making one major change for my next set of predictions: I will base many of the metrics, variables and predictive patterns on more recent fights, so as to include a shorter historical timeframe .
I will now include only the last 3 fights of each mixed martial artist. Thus, whether a fighter evolves, stagnates or devolves, the model should be able to capture that trend; for example, Tito Ortiz is not the same dominant fighter he was 6 or more years ago.
In my previous post, I mentioned I changed to the machine learning algorithm called Random Forest, a modeling technique that uses the latest and greatest mathematical methods to produce accurate results. However, a downside of Random Forest is that the results are generated in a black box; that is, one cannot see the patterns that make up the predictions. For that reason, I am reverting to a simple decision tree algorithm that allows visualization of the patterns that make up the predictions. Nevertheless, as the dataset grows to a decent size and I discover what is important, I may again use Random Forest, a more powerful technique.
Rashad Evans vs. Phil Davis
For the main event, the model has chosen Phil Davis to win based on his striking, takedown, and submission performance in his last 3 fights. One could argue that Rashad Evans has faced tougher competition, but Davis has put out some incredible fighting statistics.
Chael Sonnen vs. Michael Bisping
For the co-main event of the evening, the model has picked Chael Sonnen to win. He has superior takedown advantage, and has more accurate striking. He has also faced tougher competition. Chael’s wrestling advantage and takedown effectiveness will be too much for Michael Bisping.
Demian Maia vs. Chris Weidman
For the third fight of the night, the model has predicted a Chris Weidman victory. His takedown success rate is superior, due to his wrestling background; he also has 2 impressive submission victories in his last 3 fights. Again, one can argue that Demian Maia has fought tougher competition, but I have not yet figured out the best way to include that variable into the model’s equation.
Here are the results from last weekend’s event.
UFC on FX in Nashville, TN
Pat Barry vs. Christian Morecraft: Morecraft (Incorrect)
OVERALL ACCURACY YEAR TO DATE: 70%
Editor’s Note: This is the third in a series of fight predictions by guest columnist Jason Noriega. Jason works as an analyst at a U.S. National Laboratory. He has a keen interest in predictive modeling, and lectures around the United States. Jason also participates in predictive modeling competitions and has placed in the top 10% in a number of them using various machine learning algorithms.
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.