As more replay data from real-time strategy games becomes available, it might be possible to utilise a data-driven approach in order to streamline resource management in this type of game. This thesis studies the application of a data-driven approach in exploitative and explorative resource management in real-time strategy games. Previous work by the authors is summarised, detailing an algorithm for efficient gathering of resources. The algorithm provides an increase in the amounts of gathered resources, and is shown to be more predictable than the built-in approach used by the test bed. Furthermore, the thesis touches upon base expansion. Based on expert knowledge, 28 features that may be considered when expanding have been identified. Using feature selection methods, subsets containing 15 features are produced. A total of six different sets are tested using both artificial neural networks and decision trees. No subset shows a significant performance gain compared to the full feature set, indicating low noise of the data. The decision models using the feature sets are able to predict base expansions in replay data with a hit rate of up to 64.43%.