New algorithm for knowledge acquisition in inductive learning

NEW ALGORITHMS FOR KNOWLEDGE ACQUISITION IN INDUCTIVE LEARNING SUMMARY Keywords : Machine learning, Inductive learning, Knowledge acquisition, Entropy, Inductive learning algorithms. In general, knowledge can be gathered from an expert, from the resources related to the subject, from archives, or from observations and experiments. The process of knowledge gathering from an expert usually involves interviews, and thus is a time consuming method requiring careful and systematic examination. Although the experts can utilize their skills comfortably in their daily life, they may not demonstrate the same achievement in summarizing the knowledge, as well as making use of it within an expert system. To evaluate the available data and to translate it into an expert system database demands a different proficiency. The creation of the database through this route is time consuming and costly because of the need for the skilled human resources to accomplish the job which is a difficult task requiring higher costs. Therefore, researchers identify the knowledge acquisition as the most important element in the process of developing an expert system. Such difficulties have directed researchers to developing alternative techniques for going beyond this obstruction. Several successful procedures have been/are being developed on this subject and encouraging results have already been obtained. The aim of these newly developed techniques is to automatize the process of knowledge acquisition. In this study, three algorithms have been developed to make available the knowledge acquisition: REX-1, REX-2 and REX-3. In order to evaluate the performance of these algorithms, the database taken from real life and are employed in this field in the world have been used. Performance comparisons with today's mostly used algorithms, such as ID3, C4.5, RULES-3, RULES 3 PLUS, RULES-4, İLA, ILA-2, PRISM, OC1, CN2, J-PRUNED, GDT-NR and GDT-RS, have been carried out and the results from our algorithms were found to be acceptable. A program written in DELPHI-6 was also prepared in order to utilize these algorithms for larger databases. XIII