Prediction Result of Dota 2 Games Using Improved SVM Classifier Based on Particle Swarm Optimization

Abstract

The victory prediction of DotA 2 game is an interesting thing to know as it is one of the popular online games and is often played by players around the world. The winning opportunity of a team in this game is largely determined by the combination of heroes chosen so that a proper selection of heroes can give the team a victory. A classification is one of many ways used to make prediction. In this study, the classification will be divided into 2 classes, namely win and lose. One popular classification method is the Support Vector Machine (SVM). SVM method is suitable for classification based on 2 classes. However, in the SVM method, it is necessary to determine the optimum parameters to obtain good accuracy results, so that this research will use one of the optimization methods of Particle Swarm Optimization (PSO) for optimization of SVM parameters to increase the accuracy value. In this study, the SVM parameter to be optimized is C Parameter on linear kernel and it is proven that Optimization of C Parameter on SVM using PSO can increase SVM accuracy from 0.53866 to 0.600349718.