JSE_V2_N1_RP1 H. Bellaaj R. Ketata I. Maaloul M. Chtourou M. Ben Jemaa Journal on Software Engineering 2230 – 7168 2 1 10 23 Generating fuzzy rules, gradient descent method, merging fuzzy sets and similarity This paper presents a new algorithm for generating fuzzy rule base from numerical data. The algorithm is based on different concepts: First, the Mendel Wang generating method for constructing rules premises. Second, the gradient descent method for the identification of Takagi-Sugeno-Kang (TSK) parameters. Third, the similarity measure between fuzzy sets premises and TSK parameters. The principal idea consists in the adjustment of membership function if similarity exists using initial numerical values. The benefits consist in a better fuzzy sets definition without reducing fuzzy rule bases or losing precision. This paper focuses on the application of this approach to the non linear function and a medical problem. July - September 2007 Copyright © 2007 i-manager publications. All rights reserved. i-manager Publications http://www.imanagerpublications.com/Article.aspx?ArticleId=670