Extending a tabular knowledge-based framework with feature selection

Abstract

Tabular knowledge-based systems are known for their ease in verification and validation of knowledge bases. The main drawback of these systems is the combinatorial explosion that occurs as the number of conditions used in the table is increased. In this paper, we alleviate this problem by incorporating a new feature selection method, based on the ‘blurring' measure, in the tabular knowledge-based framework. The framework consists of three stages. In the first stage, raw data are pre-processed to reduce the data set sufficiently using feature selection. Rules are then generated and incorporated in the system. In the second stage, based on the extracted rules, the knowledge is modelled by means of decision tables. Verification and validation checks are also performed during this stage. In the final stage of the framework, the modelled knowledge is incorporated in an expert system environment, to facilitate consultation of the knowledge base. The different stages of the framework are illustrated using direct mail-order company data.