|Journal||IEEE Transactions on Knowledge and Data Engineering|
|Publication status||Published (in print/issue) - 1 Jan 2007|
Bibliographical noteOther Details
This paper investigated issues associated with machine learning approaches for enhanced decision-making performance with real-world data-sets that have mixed data types, incomplete instances, and variable feature space. With symbolic machine learning approaches, there is a requirement to transform continuous attribute values to symbolic data. The paper develops a novel distribution-index-based discretiser for such a transformation. The discretiser is combined with a rough-set theory based approach to further improve decision accuracy. Experimental results on benchmark data sets demonstrate that the approach provides superior accuracies for the same data sets compared with existing techniques. The work contributed to a successful PhD completion.