AI prediction and League of legends

Hello everyone,

I’d like to introduce you to this project I’m working on.

AI and League of legends is my project that I have been working on for a year. I study Data Science and I had the challenge to predict any LOL match. Using various opendata sources and the Riot API. I managed to design an artificial intelligence model for classification. Based on 1300 games ranging from bronze to low platinum. The model is able to give a fairly accurate and representative winrate percentage.

The test results were as follows:
Out of 200 random games in my sample, it managed to predict the win on 95% of the cases.

To deploy the model, I chose to use streamlit because I just learned it in a company and it’s great!
Although it is limited, the deployment of the model was done correctly.

Here I have taken an example of 1 game to illustrate the results.

On the one hand, I used the model to predict the team of the khazix and its opponents. In reality, the khazix had a 26% chance of winning the game and he did lose the game. His opponents had a 67% chance of winning the game and they did win. Compare the 2 percentages and see who has the higher percentage to determine who will win the game.

I let you give me feedback on my application in order to improve it. Here is the streamlit link:
My next step is to collect high elo, platinum, diamond, master, GM and challenger games!!! To do the same thing again.

The variables on which my model is based cannot be disclosed. I might write an article about it because I discovered a lot of interesting things.

For those in the know, the metrics of my model are as follows:
AUC: 0.9869
Accuracy: 95.61%.
Loss: 1.5136

I’m thinking of writing an article about this first model I designed. If you have any questions about it, I invite you to ask them below.


Can your ai model also apply to pros match (LCS,LCK,etc) too? I am looking forward to seeing you share more about the algorithm and concepts of your model! It makes me feel so excited!!

Hi, so my current model is based on low elo ranked matches (bronze to platinum). I’m currently strengthening my model on 1000 matches at any rank from iron to challenger.
For spoilers, I’ve already collected all these matches and I’m in the modelling phase.

Concerning the prediction of competitive matches (LEC, MSI etc.), I’m still in the modelling phase. It’s still difficult to determine but maybe we’ll see later. I’m not betting any matches on LOL but if I see a useful need then maybe that will be the next project.

Once the model is deployed I’ll write a short technical article about the work. That way, everyone will know how we can predict the outcome of a match.

This topic was automatically closed 180 days after the last reply. New replies are no longer allowed.