The Nomination for Evaluation of classifier performance and the impact of learning algorithm parameters by Niklas Lavesson

The thesis studies problems that are very central to AI and have deep theoretical consequences, such as "What is a good classifier?", "Is parameter tuning more important that the choice of learning algorithm?". Lavesson skillfully and with great patience studied these difficult issues using an innovative experimental approach.

The work led to a paper that has been submitted to IEEE International Conference on Intelligent Systems 2004: "A multi-dimensional measure function for classifier performance" by Niklas Lavesson and Paul Davidsson. In addition, a second paper is currently being produced: "The importance of learning algorithm parameter tuning" by the same authors.

Paul Davidsson