Knowledge Discovery: From Uncertainty to Ambiguity and Back

Abstract

Knowledge Discovery in Databases is concerned with the development of methods and techniques for making sense of data. Its aim is to model the shapes of distributions and to discover patterns. During the knowledge acquisition process choices are made. Uncertainty and ambiguity hinder the process and “poor” choices cannot be avoided. Uncertainty corresponds to situations in which the choices are unclear and/or their consequences difficult to measure. Ambiguity arises from the lack of context, there is not sufficient information to assure the success of the choice, thus causing confusion. And decision making is hampered by perceptions of uncertainty and ambiguity.

Medical diagnosis/prognosis is a complex decision making process. In this paper we present a comparative study of model ambiguity on breast cancer predictions. Automatic classification of breast cancer on mammograms using two models, logistic regression and artificial neural networks are considered. The models were trained and tested to separate malignant and benign tumors for two different scenarios. Ambiguity in the prediction was studied for the different models.

The results show that a measure of uncertainty is practical to explain observable phenomena, such as medical data. Since model ambiguity can rarely be avoided, ordering alternative models by their degree of ambiguity is crucial in medical decision making processes. Furthermore, the levels of uncertainty and ambiguity are relevant in the knowledge representation process and open up new possibilities for richer Data Mining tasks.