This thesis is a study of Magic: The Gathering card price fluctuations using
the most appropriate machine learning methods. The goal was to construct
a predictive model for card prices. This required us to identify crucial attributes,
gather necessary data, convert it to a machine-readable format and
select a suitable learning algorithm for the task. The resulting model was
effective, attaining a 61 % price trend accuracy with mythic rare cards, while
it was less successful with rare cards with only 52% accuracy, which failed to
beat default accuracy. Support vector machines algorithms and the machine
learning toolbox Weka were used to achieve these results, which were applied
in further experiments that led to the discovery of previously unknown data
dependencies.
|
