Emergence of player tactics by expert-guided machine learning: an industry tower defence case study

Modern games generate a large amount of player data that can enhance the game design and development process. Game developers can potentially utilise data science methods to extract information to inform decision making as they strive to improve the user experience and meet business goals. However, this practice is far from widespread due to the specialist expertise needed. Additionally, games are often complex, where the large number of possible player actions creates datasets with a vast state space and high dimensionality. Additionally, these player actions often require context to fully interpret and analyse. In this emerging research field, a further challenge is in ensuring proposed methods are suitable for commercial game development environments, where genres, available data sources and the production process must be considered.
This work presents the results of an industry academic collaboration, applying the less common player tactic classification method, individual sequence mining, on a fast paced, commercially available tower defence mobile game. It proposes and evaluates a novel pipeline for validating and discovering player tactics to facilitate game balancing. Rather than being applied to data captured from an analytics framework, the analysis was conducted on data captured from network messages generated by game clients. The real-time nature of these network messages creates potential for this data source to have value beyond tactics classification, with opportunities to integrate into AI pipelines for purposes such as automation.
The resulting mixed methods process demonstrates the ability of using this data source to generate insight on player tactics to game development teams, and it being feasible within the commercial game development process. The pipeline can be applied by other games companies seeking to extract value from data that is collected to make better games for their player base.