StarCraft Data Mining



The StarCraft Data Mining project is part of an effort to automatically aquire knowledge about expert RTS gameplay by mining thousands of professional StarCraft replays. By mining replays, it is possible to predict an opponent's strategy and also predict when the player will execute the strategy. The results of this project are presented in the publications section.

Histograms of Factory and Spawning Pool timings in StarCraft

Strategy Prediction Rate versus Game Time in Protoss versus Terran games



We are providing the ARFF files generated by this project to encourage other researchers to perform player modeling. Each feature vector encodes an entire game. The features represent when a specific unit or building is first produced.