An experiment in knowledge discovery using data dependencies

Abstract

The management of uncertainty has received much attention recently in the fields of database and artificial intelligence. Several methods of evidential reasoning have been proposed for real‐world problems with which uncertainty is associated. One of these problems is that of classification and it is encountered in many domains including medicine, which is considered here. Focuses on a classification technique for knowledge discovery (KD). Reasoning about classifications is a primary interest in KD. Obtains evidence to confirm or refute classes by searching for any data dependencies which exist between a classifier attribute and any of the property attributes. To illustrate the method, compares a neural network classification with one based on Tanimoto’s method. The aim was to demonstrate the approach rather than to compare the two methods of classification. After extracting the data dependency information, employs a non‐numeric evidential reasoning method to see how well this evidence supports each of the two respective classifications.