UFC AI FIGHT JOURNEY
Over the course of November our team shifted their focus towards improving our UFC Artifitical Intelligence. For anyone unaware, the UFC AI predicts the outcome of UFC fights as well as whether it thinks they will go the distance or not. Just like in Mixed Martial Arts (MMA) the AI needs training so it can learn to predict more accurately.
The AI needed training, however the focus of the training required in-depth analysis by our team. They focused on key indicators on how a fighter was going to perform whilst balancing out the availability of these indicators. For example, getting data on the height and reach of fighters is relatively simple, BUT getting whether fighters missed weight or not 5 years ago was not feasible, especially as this data could not be collected automatically.
Our team were able to identify what they believed were the key factors to influencing the outcome of fights. The most interesting factors which our AI deemed highly influencial on a fighter’s performance included their activity in the UFC and their reach. These insights could possibly be used by UFC coaches in the future (a project for the future?). Once these factors had been identified the AI could be backtested. This moved us onto the next stage of training results.
The results of our team’s training can be seen below. These results are based on the AI predicting fights from January 2019 to November 2020 and their being 1 Unit (U) placed per prediction.
As you can see the training was seemingly very successful. However just like a fight no matter the amount of training you put in, it does not mean that you will have the same results on fight night. The only way to see if the training has paid off is by getting in the cage.
GETTING IN THE CAGE
Now our team had done all the training they seemed fit, it was time to test our AI out live. Our first test was UFC 255.
On first glance, it didn’t go to plan. We had 2 wins and 5 losses meaning we lost 3.8 Units. Back to the drawing board perhaps? No. We still had our other AI predictions to assess. These predictions covered whether the fight would go the distance or not.
This is where we got creative. We decided to combine the 2 predictions together. For example one prediction had Fighter X to Win and the other prediction had Fighter X’s fight to finish inside the distance we combined them. This gave us great odds on the 1 fight at UFC 255 where this was true. This prediction was Antonina Shevchenko to win inside the distance @ 6.50 odds. Guess what, it hit! This meant we went from being -3.8 Units to + 1.7U.
After holding our breath and getting through UFC 255 we needed to ensure that this had not just been a fluke. Then came our second test, UFC Vegas 15.
Unlike UFC 255 we went positive on fight predictions alone with a record of 4 wins & 2 losses meaning we won 2.9 Units. Pretty great. Furthermore we had 2 joint predictions.
- Porter to Win inside the distance – LOSS
- Norma Dumont to win by Decision @ 3.25 – WIN
This meant for this event we went +4.15 Units. Amazing.
Based on these last 2 events we have been impressed with the performance of the AI so far. The training seems to have paid off! We acknowledge that the AI will not always be profitable on every UFC event but long term it is looking pretty great!
We will be keeping a close eye on the how the AI performs for future events to see whether all the training paid off!
Akka = The name of our AI.
ROI (Return on Investment) = Profit as a percentage of total money invested.
Calculation: (Money Won – Total Money Staked)/ Total Money Staked
Over 2 = Any of our AI’s predictions where odds were greater than 2 and less than 3 (decimal odds).
Unit = Standardisation of money (e.g. 1 unit may be £100 for one person and £10 for another). Allows better comparison of results.
Bookmaker’s Predicted Probability = Converting bookmaker odds to a probability.
Calculation (based on decimal odds): 1 / Odds
Stats Zone = Part of our Football Package offering stats for 100+ countries.
Backtest = Use the AI to predict past data showing the accuracy levels if used in the past.
Sharp = When odds are very close to the true outcome of an event meaning it is difficult to profit from these bets.