Games Engineering Programs

Formative Design and Development of a Graphical User Interface for Learning Classifier Systems

MCS Bachelor Thesis
Created by
Tim Schneider


This bachelor thesis aims to create a graphical user interface (GUI) to teach learning classifier systems (LCS), a type of machine learning. There are two major styles of LCS, Michigan, on which this thesis focuses, and Pittsburgh. A Michigan-style LCS consists of a set of classifiers called population. A classifier contains a rule, which can be thought of as a conditional statement. For example, “if the input matches the rule’s condition, then do the rule’s action.” The population is improved by learning from many inputs. When a classifier is correct or incorrect, it is either rewarded or punished, respectively increasing or decreasing its fitness, which is called reinforcement learning. Through survival of the fittest, the rule population is being improved. This means in this situation, that when the population exceeds its maximum size because additional classifiers were added, the least fit classifiers are discarded. New classifiers are introduced by combining two classifiers in the genetic algorithm. The user learns and secures knowledge about LCS in this application by experimenting and trying things out.

In order to create the learning GUI for LCS, studies were conducted with experts and novices to gather their unique insights and tailor the GUI to their needs. First, a mock-up was created to get feedback on the design quickly. After implementing the feedback and improving the design, the functional prototype with a minimal set of features was developed. Feedback was gathered, and the design was once more improved. The final application was made from this design. Then both beginners and specialists have examined the application to determine its efficacy.

Legal Information