Predicting the result of a LOL match using in game stats.

Motivation

It is that time of the year again. With the League of Legends (LOL) Worlds Champion title being up for grabs, teams from around the world are competing to see who have the better read on the meta and who would be crowned our 2020 LOL Worlds Champion. Being a group of data science students from University of Helsinki ourselves with a lot of passion in League, we set out to tackle the issue of, “Can we predict the match outcome of a LOL match using the in-game stats?”

With all the data from the games at our disposal, we proceeded with visualizing the data to hopefully gain some insights on the matter. We first plotted some stats of the number of games against the difference in kills/deaths/towers of the winning team as we deemed it would be a great indicator to predict the outcome of the match.

With these graphs we can easily see at which point a team is more likely to win. For example, in the wins vs difference of kills graph, the peaks indicate that most teams finish the games victorious at around +10 kills difference.

As we know that gold is one of the most important resources in the game as it can really facilitate the snowballing of your lead to close the game out, we had also plotted the number of games against difference of gold graph of the winning team and also a cumulative graph to better see where the difference of gold really makes a difference.

With the cumulative graph, we notice that more than half of the teams win with a gold lead of ~8000.

In addition, we examined the percentage of the winning teams’ first objectives control. They are first blood, first tower, first dragon, first baron and first inhibitor.

From the graph we find that the winning teams have an uncommonly high percentage of securing the first tower at around 77%. We suspect that this is due to the nature of the first tower giving more gold to the players thus allowing them to gain a lead earlier over their opponents. Hence, we encourage the players to play around ways of taking down the first tower, e.g. taking the rift herald to help take down the first tower.

Finally, we looked at the number of dragons and barons of the winning team. Similar to before, we also plotted the cumulative percentage of wins against the number of dragons/barons taken. In this way, we could tell from what amount of dragons or barons you are more likely to win.

We see that taking the second dragon drastically increases the chance of you winning the game. More than 60% of games were won after taking two or more dragons.

We were able to achieve a testing accuracy of 98% using all of the aforementioned features. We also found out what our model deemed were the most useful features/objectives in determining if you have won/lost a game.

In this article, we have visualized some of the most important stats in the game and also showcased the usage of our machine learning models to get a live probability of your game being a win using the end-game data from the Challenger League. However, we hope to extend our model to include the live game stats in the future.

Related posts:

A 20-year yoga journey that started with a message from Madonna, dreams of a judgment-free world, and lead to finding my own path.

On the week of Wonder Woman Day 2020— October 21st — and having reached four months at my new role working with a well-funded fintech startup that focuses on Amazon’s cloud technology, I find myself…

Our current classroom experience is very generic. A video for students to watch and a chat for them to interact(very much like youtube’s Live Videos) The task is to think of an end to end classroom…