Victory prediction from League of Legend champion selection data

• Victory prediction from League of Legend champion selection data
  • Objectif
  • Dataset description
    • Splits
    • Files
      • Champions
      • Matches
  • Getting started

Objectif

The continuous development of e-sports is generating a daily trail of insightful data in high volume, to the point that justifies the use of exploratory data analysis.

In particular, the multiplayer online battle arena (MOBA) game League of Legends (LoL), organizes one of the most viewed tournaments, attracting over 4 million peak viewers.

The game lets participants choose between more than 161 champions with different characteristics and game play mechanics affecting the dynamics of team composition. Thus, champion selection is of capital importance for pro players.

Multiple works focused on champion selection data in order to predict team victory for DOTA, a MOBA similar to League of Legends, but LoL is still under-researched. And with the regular new patches received, it is difficult to compare predictor performances across time.

To this objective, we are releasing this curated dataset such that others can try their own architectures on victory prediction from champion selection data, thus offering a benchmark dataset for the community.

Dataset description

This dataset has been collected by Devoteam Revolve from Riot Developer API

http://france.devoteam.com/wp-content/uploads/sites/21/2021/05/logo-cartouches-RVB-ROUGE.png" alt="Devoteam logo">

The dataset has a total of 84440 games that are from 2022 at the version 12.12 of the game.

The games are only from the highest ELO players, with ranks of either Master, Grand Master and Challenger. This ranks represents the top 1.2% of all players.

Splits

The dataset comes pre splitted

Files

Dataset organization:

12.12.-splits
├── test
|  ├── df_00000.csv
|  |   ...
|  └── df_xxxxx.csv
|
├── train
|  ├── df_00000.csv
|  |   ...
|  └── df_xxxxx.csv
|
└── val
|  ├── df_00000.csv
|  |   ...
|  └── df_xxxxx.csv
|
└── champion.json

Champions

All champions information can be found under ./12.12.-splits/champion.json

This file allows the conversion from Player_{Player_id}_pick id number to the champion name.

Multiple other information are also freely available such has champion damages, HP, etc ...

Matches

All the matches are collected in the 3 directories:

• ./12.12.-splits/train/
• ./12.12.-splits/val/
• ./12.12.-splits/test/

Each of these directories contain multiple df_xxxxx.csv files detailing up to 100 matches.

The description of each column can be read in the below table.

The column which possess {Player_id} in their name are repeated 10 times, one for each player.

For example, the column name Player_{Player_id}_team can be found in each csv as 10 different columns with names ranging from Player_1_team to Player_10_team.

2019

This dataset comprises comprehensive information from ranked matches played in the game League of Legends, spanning the time frame between January 12, 2023, and May 18, 2023. The matches cover a wide range of skill levels, specifically from the Iron tier to the Diamond tier.

The dataset is structured based on time intervals, presenting game data at various percentages of elapsed game time, including 20%, 40%, 60%, 80%, and 100%. For each interval, detailed match statistics, player performance metrics, objective control, gold distribution, and other vital in-game information are provided.

This collection of data not only offers insights into how matches evolve and strategies change over different phases of the game but also enables the exploration of player behavior and decision-making as matches progress. Researchers and analysts in the field of esports and game analytics will find this dataset valuable for studying trends, developing predictive models, and gaining a deeper understanding of the dynamics within ranked League of Legends matches across different skill tiers.

Released in October 2020, League of Legends: Wild Rift is a mobile MOBA (multiplayer online battle arena) gaming title that was developed and published by Riot Games. The title is a mobile version of the MOBA genre classic League of Legends and generated approximately 680,000 app downloads in June 2025.

In June 2025, League of Legends: Wild Rift generated approximately 5.96 million U.S. dollars in in-app purchase revenues. The multiplayer online battle arena (MOBA) gaming title is a mobile version of the PC game League of Legends, published by Riot Games.

This dataset contains up-to-date information on the characteristics (e.g., roles, stats) of every champion in League of Legends as of Season 15, patch 25.05. It was scraped from the League of Legends Wiki Champion Data Module on March 14th, 2025.

The source of the headline image is Andrei Castanha on Unsplash.

Click the remote source link to see the script used to scrape the data.

Note that the dataset provides stats for all game modes (e.g., SwiftPlay, ARAM).

Related datasets: - 25.S1.1 League of Legends Champion Data (2025) - 25.S1.3 League of Legends Champion Data (2025) - 25.S1.4 League of Legends Champion Data (2025) - 25.09 League of Legends Champion Data (2025) - 25.011 League of Legends Champion Data (2025)

Dataset Description

Disclaimer

This work isn’t endorsed by Riot Games and doesn’t reflect the views or opinions of Riot Games or anyone officially involved in producing or managing League of Legends. League of Legends and Riot Games are trademarks or registered trademarks of Riot Games, Inc.

  League of Legends Replays Dataset

This dataset contains over 1TB+ (100k+ replays) of League of Legends game replay data for research in gaming analytics, behavioral modeling, and reinforcement learning… See the full description on the dataset page: https://huggingface.co/datasets/maknee/league-of-legends-decoded-replay-packets.

League Of Legends Icon Detection

## Overview

League Of Legends Icon Detection is a dataset for object detection tasks - it contains League Icons annotations for 358 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

This dataset contains start & end timestamps for League of Legends games played via the Duowan plugin. We randomly sampled 100k user_ids and gathered all games associated with those users.

Gold Bar League Of Legends

## Overview

Gold Bar League Of Legends is a dataset for object detection tasks - it contains Gold Bar annotations for 673 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

Real-time player count data for League of Legends across PC and mobile platforms

In competitive MOBA game League of Legends, a ranking system is then used to match players of similar skill levels against each other in online games. The League of Legends ranking system has ten levels to indicate the skill level of the gamer. Among North American players, the Bronze skill level was the most common tier as of January 2025, achieved by 21.63 percent of players. Meanwhile, only a small handful of gamers were able to reach the top levels within the game. For casual and professional gamers alike League of Legends has built a loyal and large fanbase in the ten years since its release. The game reached a landmark 100 million monthly active users in 2016, an impressive increase from the 15 million users it had in 2011. As a result of its entertaining and varied gameplay, League of Legends has also become one of the most popular games amongst eSports viewers. The League of Legends eSports championship finals in 2023 became one of the most watched eSports events in history as as viewership peaked at 6.4 million viewers. With so many people watching this event, it comes as no surprise that the prize pool is sizeable. The 2023 edition took place in the Seoul and Busan, South Korea, with a total of 2.23 million U.S. dollars in prize money up for grabs. The South Korean eSports team T1 emerged victorious in 2023, thereby pocketing a whopping 445 thousand U.S. dollars in prize money.

League of Legends is a free online battle arena game. The objective of this game, in almost all game modes, is to destroy the enemy Nexus. The game is not only popular among players, but also among eSports viewers. League of Legends events on the video streaming service Twitch were watched for a combined 78.9 million hours in April 2025.

英雄联盟(League Of Legends,LOL)

## Overview

英雄联盟(League Of Legends,LOL) is a dataset for object detection tasks - it contains Legend Minion Turret annotations for 1,154 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

League of Legends Match Data

A comprehensive dataset collection and processing system for League of Legends match data using the Riot Games API.

  📊 Data Structure





  Match-Level Fields

match_id: Unique match identifier game_duration: Match duration in seconds queue_id: Game queue type (filtered to 420 for ranked solo/duo)

  Player-Level Fields





  Basic Information

summoner_name: Player's summoner name summoner_id: Unique summoner identifier… See the full description on the dataset page: https://huggingface.co/datasets/BoostedJonP/league_of_legends_match_data.

Dataset

This dataset was created by Danilo Donato

Contents

It contains the following files:

League of Legends Skin Splash Art Collection

A comprehensive, high-resolution dataset of splash art for every skin in League of Legends, organized by champion.

Context

League of Legends, developed by Riot Games, is one of the world's most popular video games, renowned not only for its strategic gameplay but also for its rich lore and stunning visual artistry. A core part of the game's aesthetic appeal lies in its vast collection of character skins, which reimagine champions in different themes and universes. Each skin is introduced with a unique, high-quality illustration known as "splash art."

This dataset aims to be a comprehensive and up-to-date collection of these splash art images, providing a valuable resource for data scientists, machine learning engineers, and fans of the game.

Content

The dataset contains high-resolution splash art images for every skin in League of Legends. The data is organized into a clean, hierarchical structure for ease of use.

The file structure is as follows:

• A main directory contains subdirectories for each champion.
• Each subdirectory is named after a League of Legends champion (e.g., Ahri, Garen, Jinx).
• Inside each champion's folder, you will find the splash art for all of their respective skins.
• Each image is a high-resolution .jpg file named after the full skin title (e.g., Star Guardian Ahri.jpg, God-King Garen.jpg).

Example Structure:

/
├── Ahri/
│  ├── K-DA Ahri.jpg
│  ├── Spirit Blossom Ahri.jpg
│  └── ...
├── Hecarim/
│  ├── Nightbringer Hecarim.jpg
│  ├── Arcade Hecarim.jpg
│  └── ...
└── ... (and so on for every champion)

Acknowledgements

This dataset was created by scraping the public-facing website lolskin.info. A huge thank you to the creators and maintainers of lolskin.info for curating and providing this information and imagery to the community.

Please note that this is a community-created dataset. League of Legends and all its assets are the intellectual property of Riot Games, Inc.

Inspiration & Potential Use Cases

This rich visual dataset is perfect for a wide range of computer vision tasks. Here are a few ideas to get you started:

• Image Classification: Train a model to classify skins by champion, skin line (e.g., 'PROJECT', 'Star Guardian', 'K/DA'), or even rarity.
• Color Palette Analysis: Analyze the dominant color palettes across different skin themes or champions. Are 'dark' themes like 'Nightbringer' measurably different from 'light' themes like 'Dawnbringer'?
• Similarity Search: Build a model that can find visually similar skins. If a user likes 'PROJECT: Vayne', what other skins might they like?
• Object Detection: Can you train a model to reliably detect the champion within the often chaotic and detailed background of the splash art?

Note the scraper was ran on 6/30/2025 1:00PM PST. in order to update the script please use node to run the script below:

const fs = require('fs/promises');
const path = require('path');
const fetch = require('node-fetch');
const cheerio = require('cheerio');
const puppeteer = require('puppeteer');

// --- Configuration ---
const BASE_URL = 'https://www.lolskin.info';
const START_URL = `${BASE_URL}/en-us`;
const OUTPUT_DIR = path.join(_dirname, 'skins'); // Creates a 'skins' folder
const DELAY_BETWEEN_REQUESTS_MS = 200; // Delay to be nice to their server

/**
 * A utility function to add a delay.
 * @param {number} ms - The number of milliseconds to wait.
 */
const sleep = (ms) => new Promise(resolve => setTimeout(resolve, ms));

/**
 * Sanitizes a string to be a valid filename or directory name.
 * @param {string} name - The original string.
 * @returns {string} A safe name.
 */
function sanitizeFilename(name) {
  return name.replace(/[\?%*:|"<>./]/g, '-').trim();
}

/**
 * Downloads an image from a URL and saves it to a specified path.
 * @param {string} imageUrl - The full URL of the image to download.
 * @param {string} filePath - The local path to save the image to.
 */
async function downloadImage(imageUrl, filePath) {
  try {
    const response = await fetch(imageUrl);
    if (!response.ok) {
      throw new Error(`Failed to fetch image: ${response.statusText}`);
    }
    const buffer = await response.buffer();
    await fs.writeFile(filePath, buffer);
    console.log(`   ✅ Saved: ${path.basename(filePath)}`);
  } catch (error) {
    console.error(`   ❌ Error downloading image for ${path.basename(filePath)}:`, error.message);
  }
}

/**
 * Uses Puppeteer to scroll to the bottom of the page to load all content.
 * @param {import('puppeteer').Page} page - The Puppeteer page object.
 * @param {string} selector - The CSS selector for the items being loaded.
 */
async function scr...

LoL Object Detection

## Overview

LoL Object Detection is a dataset for object detection tasks - it contains Objects annotations for 2,988 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [MIT license](https://creativecommons.org/licenses/MIT).