Peter J. Deer

A study of parameter values for a Mahalanobis distance fuzzy classifier

A supervised Mahalanobis Distance fuzzy classifier (and the related fuzzy c-means clustering algorithm) requires the a priori selection of a weighting parameter called the fuzzy exponent. Guidance in the existing literature on an appropriate value is not definitive. This paper attempts to rigorously justify previous experimental findings on suitable values for this fuzzy exponent, using the criterion that fuzzy set memberships reflect class proportions in the mixed pixels of a remotely sensed image.

Toward cooperatively-built knowledge repositories

After noting that informal documents and formal knowledge bases are far from ideal for discussing or retrieving technical knowledge, we propose mechanisms to support the sharing, re-use and cooperative update of semi-formal semantic networks, assign values to contributions and credits to the contributors. We then propose ontological elements to guide and normalize the construction of such knowledge repositories, and an approach to permit the comparison of tools or techniques.

A new approach on ρ to decision making using belief functions under incomplete information

This paper discusses an expected utility approach on ρ to decision making under incomplete information using the belief function framework. In order to make rational decisions under incomplete information, some subjective assumptions often need to be made because of the interval representations of the belief functions. We assume that a decision maker may have some evidence from different sources about the value of ρ, and this evidence can also be represented by a belief function or can result in a unique consonant belief function that is constrained by the evidence over the same frame of discernment. We thus propose a novel approach based on the two-level reasoning Transferable Belief Model and calculate the expected utility value of ρ using pignistic probabilities transformed from the interval-based belief functions. The result can then be used to make a choice between overlapped expected value intervals. Our assumption is between the strongest assumption of a warranted point value of ρ and the weakest assumption of a uniform probability distribution for an unwarranted ρ.

Selecting parameter values for mahalanobis distance fuzzy classifiers

The fuzzy c-means clustering algorithm, and a related supervised classifier, require the a priori selection of a weighting parameter called the fuzzy exponent. This paper investigates suitable values of this fuzzy exponent using the criterion that fuzzy set memberships reflect class proportions in the mixed pixels of a remotely sensed image.