Regarding the machine learning comments:
I’ve been playing with some simple ML algorithms (mainly linear regression and other simple model) to predict various results. It’s not really useful for a single league, since the data sets are too small. You can come to some good conclusions, but nothing you couldn’t already figure out (getting first baron increases you chances to win, etc…).
Comparing the results across regions is pretty interesting. For example LCK teams have higher win rates after securing first turret, and especially first baron. This could indicate the LCK teams are better at pushing an advantage and winning the game by objective control.
Machine learning is a great tool for some problems. For example, given a set of champions, you could estimate a team’s probability to get first turret. In the context of betting, it’s not as useful, since there are way too many random factors, mainly picks/bans, and the constantly changing meta.
If you want to get started with machine learning, I think you should learn about how linear regression works first, then try applying it to a simple problem. Something like “if X team gets first blood/turret/baron, what is the chance of winning”?
If you just want to bet, the kind of analysis I demonstrated in this article is more than enough.
Regarding your comment about average game time and kills… this is pretty tricky. While some teams tend to have more “bloody” games than others, I’ve found it pretty tricky. The choice of champions, among other factors, is very difficult to factor in. Same with game time. I’m still exploring different techniques, so I might take a look at some other markets in a future post.