Evaluation of Clustering Methods on Game Play Data

The goal of the project is to evaluate several clustering algorithms on players’ styles data in the context of Video
Lottery Terminals (VLTs). The previous work has shown that by segmenting anonymous player data by
sessions, and then clustering the sessions using the simple k-means algorithm, we can get a descriptive
statistic on player styles, including problem gambling behavior, recreational player styles, and similar. An open
question is whether the preprocessing techniques were optimal for this purpose and whether the k-means
algorithm is the most appropriate algorithm. In this project, of a number of clustering algorithms, such as
partition-based (e.g., k-means), hierarchical-based (e.g., hierarchical k-means), density-based (e.g.,
DBSCAN), model-based (e.g., statistical model based such as EM), and grid-based (e.g., STING) algorithms
will be evaluated and their performance on game play data will be analyzed.

Soheil Latifi
Faculty Supervisor: 
Vlado Keselj
Nova Scotia
Partner University: