Generation of Rule-Based Expert Systems with Certainty Factors from Datasets

Konstantinos Kovas, Ioannis Hatzilygeroudis

Abstract: We present an approach for extracting rules from datasets using Certainty Factor values to represent uncertainty and generating Expert Systems. This paper focuses on the different methods we use for extracting rules from datasets with nominal values, the way we calculate the Certainty Factors and also our approach for combining conclusions from different sources. Furthermore, we briefly present a tool we have developed that uses these methods to semi-automatically generate Expert Systems coded in the CLIPS language. Finally, we present experimental results that show the approach to have comparable performance in the classification task with other popular methods like decision trees.