Data Analysis on UFC Matches
The skills the author demoed here can be learned through taking Data Science with Machine Learning bootcamp with NYC Data Science Academy.
About UFC
Data says the Ultimate Fighting Championship (UFC), idealised by the Gracie Family, was inspired by countless exhibition matches termed the “Gracie Challenge”, witch was an open invitation issued by some members of the Gracie Family to martial artists of other styles to fight them in a “Vale Tudo” (anything goes) match.
The purpose of these challenges was to prove the effectiveness of the Gracie style of Brazilian Jiu-jitsu. The matches typically featured a smaller Gracie versus a larger and/or more athletic looking opponent. The Gracies defeated martial artists of many different styles such as boxing, judo, karate and wrestling, while experiencing few losses.
Data Results
Looking at the initial UFC results, we can see how well Brazilian Jiu-jitsu fighters performed. 
The data shows that, Grapplers (Judo, Wrestler and Brazilian Jiu-jitsu practitioners) were superiors. It means that Striking Athletes were not able to face a Grappler properly.
But, around 2010 the "Win Ratio" per style, seems quite similar.
Now most of fighters recognises that it is essential to learn how to be safe in different aspects of fighting. The consequence of this new understanding of "how to fight" is the similarity between Grapplers and Striking Athletes's "Win Ratio".
Data Analysis of the Victories
Now, take a look at the winners:
We can see that most of them are Grapplers.
Knowing that Grapplers tends to apply Submission techniques, the table shows that something has changed in the Grappler's way to fight. They are not finishing fights using Submissions as they were before.
Current Distribution:
Distribution until 2000:
Conclusion
These insights about Strikers and Grapplers point to the fact that both groups are training more than one style.
The next questions that I'll try to answer will be:
1 - Which styles combined has the best results?
2 - Are there Athlete's characteristics that could be used to build a predictive model?
About Author
