Constructing ANFIS With Sparse Data Through Group-Based Rule Interpolation: An Evolutionary Approach

An adaptive-network-based fuzzy inference system (ANFIS) offers a popular and powerful mechanism. As with many other advanced data-driven techniques, developing an effective ANFIS typically requires sufficient training data. However, in real-world applications, it is not always straightforward to obtain large amount of representative data that cover the entire problem space accomplish required training, seriously restricting performance learned ANFIS. This article introduces new learning approach through evolutionary process, which able generate only small certain region, by interpolating well-trained ANFISs neighboring regions. Such process works first producing initial population candidate rules region shortage, rule dictionary constructed from trained neighborhood The crossover mutation operations over these are then executed effort attain candidates improved performance. When this genetic terminates, chromosomes final either collectively form or each individually represents ANFIS, depending on whether single set representing implemented chromosome within evolving population. Comparative experimental evaluations both synthetic datasets carried out, demonstrating spite proposed interpolation produce models significantly outperform those using existing mechanisms.