This diploma thesis has two main objectives; the creation of predictive models used for predicting of outcomes of professional matches played in the videogame League Of Legends, and the enhancement of professional match datasets using additional data sources. The first part of this thesis is dedicated to the act of web scraping of selected sources in order to get professional players' in-game nicknames. Data enhancement is based on three main data sources. 77 % of all professional players&a... show full abstractThis diploma thesis has two main objectives; the creation of predictive models used for predicting of outcomes of professional matches played in the videogame League Of Legends, and the enhancement of professional match datasets using additional data sources. The first part of this thesis is dedicated to the act of web scraping of selected sources in order to get professional players' in-game nicknames. Data enhancement is based on three main data sources. 77 % of all professional players' nicknames from the LEC, LCK, LCS and LPL leagues is obtained from these sources. Data from professional and recreational matches is aggregated and transformed for use in modeling. The methods of random forests and k-Nearest Neighbors are compared using different approaches to partitioning data for test purposes. Using certain test criteria and random forests, a model is obtained which is able to correctly determine the outcome of 54,67 % of matches. A random forest model which does not use a test dataset is able to correctly determine the outcome of 57,72 % of matches from the same timeframe. Analysis of grouped variable importance is conducted, during which two groups with the highest importance are identified – one such group is the player statistics in a specific league, the other is player statistics in solo queue. Analyses and transformations are carried out in Python programming language. |