Fantasy Fights

Intelligent, unique MMA analysis

SILVA 1.1 to (Hopefully) be Released Soon

While I really enjoy watching, reading about, and writing about fights, I have to admit that the recent UFC schedule has been pretty nuts. We’ve just gone through UFC events on four consecutive weeks, and there was a Strikeforce event on the week before that. So for me, that means a lot of fights to enjoy watching, but at the same time, instead of doing things like original research and working on SILVA, I have to devote my time to processing SILVA scores and researching and writing about fights.

Now, I have three weeks to do some of these other things before the next UFC event. This means that I finally have some time to address some of the problems that I’ve identified with SILVA. I have a number of ideas on ways to potentially improve the system that need to be tested, and now I’ll have the time to actually perform these tests and refine the system a little bit.

I think there’s a reasonable limit to how accurate a rating/prediction system can be for the UFC, and my educated guess is that the limit is somewhere between 63% and 65%. Right now, SILVA is good for about 61% of fights, which is a respectable success rate. What excites me is that I believe there are some issues in the system that are easily identified and correctable, meaning that there’s plenty of room to improve on the concept behind it.

Here are some ideas that I have for improving SILVA:

-The Fight Level statistic is currently based on a fighter’s last ten fights. The number of fights eligible for the statistic is about right: fights before that have relatively little relevance. For example, Ken Shamrock shouldn’t be evaluated as a fighter right now based on beating Bas Rutten in 1995. I’ve done a little bit of experimenting with how many fights should be eligible, and ten seems to be about right. The problem is what happens when a great win or bad loss goes just outside of the ten-fight window. Melvin Guillard is a perfect example. Guillard had a SILVA score of 31.54 going into his fight against Joe Lauzon. After losing to Lauzon, Guillard’s SILVA score actually increased to 33.21. The reason is that Guillard’s worst loss, to Rich Clementi, now happened 11 fights ago instead of ten. What I want is for the Fight Level statistic to slowly fade older fights out, instead of fully considering them at one moment and completely discarding them the next. If my concept works right, Guillard’s SILVA score would have slowly gone up by winning his fights recently, and then would have gone down after losing to Lauzon.

-One thing that bugs me about the current system is that it’s disjointed. That means that fighters who have a rAP above 43.00 have a SILVA score that is COMPLETELY based on their rAP. Fighters whose rAP is between 10.00 and 43.00 have a SILVA score that is COMPLETELY based on their Fight Level, NOT their rAP. And fighters whose rAP is lower than 10.00 have a SILVA score that is COMPLETELY based on their rAP. I want to find one elegant formula that combines these statistics instead of applying them in such a disjointed manner. Doing so would help to reduce the SILVA scores of fighters like Paul Sass, Edson Barboza, Chris Weidman, and Matt Mitrione to a more reasonable level, as SILVA is probably a little too enthusiastic about them at the moment.

-I’ve documented that SILVA tends to struggle with once-great fighters currently on the decline. Fighters like Fedor Emelianenko, Matt Hughes, and Takanori Gomi spent so much time winning fights and building up a huge rAP, that their rAP doesn’t decline nearly fast enough to catch up to the actual decline of their respective skills. My idea is to have rAP measure a fighter’s last 10 to 20 fights instead of his entire career.

-I want to diminish the impact of a fighter having one great win after a large number of irrelevant losses, or one bad loss after a large number of great wins.

There are other aspects of things I want to change that refer to the specific formulas I use, and I’m not prepared to disclose those at this time.

Hopefully I didn’t confuse you too much with everything written above, but my plan is to have a new and improved SILVA by the time the week of UFC 137 begins. I plan on writing things here on this blog in the meantime as well, but if it seems like content has slowed down considerably, it’s probably because I’m focused on improving SILVA. Stay tuned!


2 responses to “SILVA 1.1 to (Hopefully) be Released Soon

  1. howard morton October 11, 2011 at 9:15 am

    There are people at that have accurately predicted over 500 fights at a 75% accuracy rate. I am 8-2, 8-2, 8-3 over the last 3 ufc events. I went 10-2 at ufc 129 and made a nice chunk of change.

    I guess my point is that SILVA will never be that good consistently because it does not have “gut feeling”. But if I am wrong I will start using SILVA to make even more money.

  2. BlackSwan October 12, 2011 at 10:16 am

    If The Fight Level statistic is currently based on a fighter’s last ten fights, why not take into account the time between fights rather than the number of fights ?

    If a fighter has had 10 fights in the past 3 years and is compared to a fighter that has had 10 fights in 6 years there is going to be a disparity no ?

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: