TEA-modeling: Testing effective algorithms for modeling

• Main Authors: Ang, John Patrick S., Bagay, Andrew L., Baysa, Jude Alexis T., Po, Bryan G.
• Format: text
• Language: English
• Published: Animo Repository 2009
• Subjects: Data Mining; Reinforcement learning; Intelligent control systems; Computer adaptive testing; Computer Sciences
• Online Access: https://animorepository.dlsu.edu.ph/etd_bachelors/10710
• Institution: De La Salle University

This research aims to evaluate Machine Learning (Reinforcement Learning and Best-Response Learning Algorithm) and Data Mining algorithms (Classification, Association, and Neural Network) in terms of providing rationality and human believability in an agent. Rationality considers the time, cost and space used up in reaching the goal. It concerns mainly on making the agent the best player of the game. Human believability, on the other hand, considers how an agent manifests human-like behavior as it competes with a player or an agent. It concerns mainly on fooling human players into thinking the agent is also a fellow human. An existing snake game from a previous research shall be used as a test bed to deploy and evaluate the agents. The testing process for the rationality aspect of the agent will be based on a methodology previously researched by John C. Duchi and John E. Laird (2000), while the believability aspect will be based on a research by Christian Bauckhage, et al. (2007). In the said methodology, rationality is hugely based on win-and-loss of the agent while believability is hugely based on observation ratings on how human-like the agent is.