Bachelor Thesis: Automatic Classification of Poker Player Types from Hand Histories

 

 

Background & Motivation

 

Classifying poker players into behavioral types—such as tight-aggressive, loose-passive, or maniac—is a common practice among human professionals and a key component of exploitative strategy. However, existing research relies heavily on manual tagging or heuristic thresholds for features like VPIP (Voluntarily Put Money in Pot) or PFR (Pre-Flop Raise) [1, 2]. Automating this classification could allow for real-time adjustments and pave the way for intelligent adaptive agents in multi-agent poker settings.

 

Research Question

 

Can we develop a machine learning-based system that automatically classifies poker players into standard behavioral types using only hand history data?

 

Related Work

Existing research has applied clustering techniques to define player types based on behaviors, such as in Identifying Player’s Strategies in No Limit Texas Hold’em Poker through the Analysis of Individual Moves [1], or used reinforcement learning with opponent-model-aware reward shaping [2]. Frameworks like HoldemML aim at generating agents by imitating top human strategies [3], while ensemble classifiers have been applied for action prediction but not explicit typology labeling.

 

Objectives

 

The thesis will design, implement, and evaluate a classifier capable of assigning player types based on hand-level behavioral patterns. Optionally, the classifier can be integrated into an online system for dynamic opponent modeling.

 

If you are interested, please reach out to Alexander Studt (studt@teco.edu).

References

  1. Teófilo, L. F., & Reis, L. P. (2013). Identifying Player’s Strategies in No Limit Texas Hold’em Poker through the Analysis of Individual Moves. arXiv preprint. https://arxiv.org/abs/1301.5943 

  2. Teófilo, L. F., Passos, N., Reis, L. P., & Cardoso, H. L. (2012). Adapting Strategies to Opponent Models in Incomplete Information Games: A Reinforcement Learning Approach for Poker. In Lecture Notes in Computer Science (Vol. 7326). https://doi.org/10.1007/978-3-642-31368-4_26 

  3. HoldemML: A framework to generate No Limit Hold’em Poker agents from human player strategies. In 6th Iberian Conference on Information Systems and Technologies (CISTI 2011).