The UFC MMA Predictor https://ufcmmapredictor.herokuapp.com/ is a web app to predict fights exclusively in the Ultimate Fighting Championship (UFC). Please use at your own discretion as there is no guarentee of profit. The machine learning model behind this web app currently gives a 70.4% accuracy. This is also considered a professional project of mine. Please read the Jupyter Notebook documentation for more detailed information on the data preparation, modelling, and visualization.
Here you will find the following documents:
- Jupyter Notebook of data science pipeline for the project
- Python scripts to the web app
- R web scraping scripts
- Demo of the app
Here are a few information you need to have before using the web app:
- Fight card information (who is fighting who)
- Fighter information (weightclass and name)
- Betting odds (preferred decimal odds), knowing who is the favourite and underdog fighter beforehand
Here are some of the trends that have been spotted in recent MMA years to note:
- Picking favourite only gives 60% chance of winning in long term
- Fights go on longer in general
- Southpaws win 53% of their fights, as opposed to 48% for orthodox fighters
- Fighters who switch stances have an even higher success rate than both orthodox and southpaw fighters, with a combined win percentage of 57%
Let's take the main card of UFC 217 to see how well the app performs. Note that UFC 217 fight data is neither in the training or test set. The fights that happened were (winners in bold):
- GSP (@2.10) vs Michael Bisping (@2.00)
- Cody Garbrandt (@1.57) vs TJ Dillashaw (@2.75)
- Joanna (@1.20) vs Rose Namajunas (@7.00)
- Stephen Thompson (@1.57) vs Jorge Masvidal (@2.57)
The web app predicted all the above fights correct except for Joanna vs Rose Namajunas where Joanna had most fight stats dominant over Rose and was the big favourite over Rose. However, Rose caught Joanna with a fight ending hook during the fight and it was indeed 'Goodnight Irene'.
I would appreciate it if you cite my Github repo if you ever use any information or derived any inspiration from here.