Feature Extraction for StarCraft II League Prediction & Feature Construction combining GP and PSO

Abstract

The players are divided into seven leagues depending on their skill in StarCraft II. Since it it most ideal for players with similar skills to play games, it is important to be included in the appropriate league. In our research, we compared features extracted from several sections of replays and proposed a method combining genetic programming and particle swarm optimization for feature construction and selection. Comparing the features, the longer the section, the higher the accuracy. Our proposed method showed higher accuracy than standard multiple feature construction method. As a result, we classied the player's league with about 67% accuracy.