SFIST-based Fast Data Classification

Abstract

Fuzzy logic is a powerful tool in computer science, which has been used in countless applications since its conception in the late 80s. Numerous classifiers have been based on it, taking advantage of the flexibility and robustness against noise that is inherent in fuzzy systems. One such classifier called the “Sequential Fuzzy Indexing Tables Classifier” has been developed, to provide a fast and robust classification performance by combining the speed of indexing tables with the flexibility of fuzzy inference systems. One major disadvantage of it is its memory requirement that scales exponentially with the dimension size of the problem. To solve this problem, the authors have proposed the so-called Sequential Fuzzy Indexed Search Trees (SFIST) classifier that uses the same principle, but with a much smaller structure. In previous works, the authors have proposed two variants for the SFIST classifier, and both were shown to drastically reduce the required memory space compared to that of its predecessor, without any loss in classification performance. In this paper, a new, third variant is proposed that implements a hybrid approach between the first two, aiming to further improve the classification accuracy, without sacrificing too much operational speed.