Summary : We were able to attain 71.3% pre-game accuracy as to whether a team would win against another team. This accuracy goes up marginally as the match progresses as more data is attained. Full Report Here : https://github.com/lppier/League_Of_Legends_Data_Analytics/blob/master/FT01_LeagueOfLegends_Report.pdf
Data Analytics on League of Legends dataset from Kaggle : https://www.kaggle.com/chuckephron/leagueoflegends We assumed the role of investigating odds of a certain team winning in an e-sports situation at certain time points. (before the game, 5 minutes into the game, 10 minutes into the game and halfway through the game).
This was a group project in data mining following the CRISP-DM methodology :
- Business understanding
- Data understanding
- Data preparation
- Modeling
- Evaluation
- Deployment
Personally, I did some data pre-processing, creating new predictors out of exisiting features. I also ran the models in R to investigate the accuracy at various time slices. For the report, I worked on the interpretation and evaluation of the models' metrics. eg. ROC curve, confusion matrix, etc.
Anurag Chatterjee
Bhujbal Vaibhav Shivaji
Charles Thomas De Leau
Lim Pier
Liu Theodorus David Leonardi
Tsan Yee Soon
Some work in R was done to transform data from the original fields to fields that we felt were valuable as predictors for modeling. eg. win efficiency
We investgated random forests, xgboost, neural networks, decision trees and logistic regression over the course of this project. The eventual model chosen was logistic regression.
The R Markdown files (cleaned up) are included. This repository is the final one that contains the submitted findings. Working repository during the project : https://github.com/lppier/KE5107_LeagueofLegends