Real-Time VLT Player Data Personae Classification

The goal of this project is to train a machine learning model that can identify player’s personae using VLT (Video Lottery Terminal) data within a transactiontime limit. The personae are results of the previews MITACS project. Using unsupervised learning each playing session was associated with a playstyle. Identifying the playstyle as soon as possible is of great importance since it can be used later for problem gambling detection in early stages of playing. Another challenge this project tries to solve is to estimate an optimal limit for transactiontime. It is preferred to identify the personae earlier with higher accuracy. Another challenge is to propose a good method for feature extraction. For the previews MITACS project a complete session was used for identifying the personae, therefore the feature space is not sparse unlike this project. To overcome sparse feature space problem, proposing new features or deep neural networks that ease feature extraction task.

Faculty Supervisor:

Vlado Keselj

Student:

Soheil Latifi

Partner:

IGT

Discipline:

Computer science

Sector:

University:

Dalhousie University

Program: