Project Details

Layman's description

Research and development of methods for evidence evaluation for skewed data and a comparison of AI approaches for dimension reduction with statistical approaches.

Key findings

Likelihood ratios for two-level models were determined with the use of a Gaussian kernel estimation procedure when the between-group distribution was thought to be non-normal. Work on the project developed a two-level model for the case in which the between-group distribution is very positively skewed and an exponential distribution may be thought to represent a good model.

Work was undertaken in the area of feature selection from the AI perspective, with three papers arising from the work. The AI approach to feature selection was in contrast to a statistical approach described in another paper.

Three new approaches to fuzzy-rough feature selection based on fuzzy similarity relations were also proposed. In particular, a fuzzy extension to crisp discernibility matrices was proposed and utilized. Initial experimentation showed that the methods greatly reduce dimensionality whilst preserving classification accuracy.

Another output presented an outline of the basic concepts of rough sets and their major extensions, covering variable precision, tolerance and fuzzy rough sets. It also showed the diversity of successful applications these theories have entailed.
Effective start/end date1/10/0431/03/07


  • EPSRC: £90,598.00