A comparative experimental evaluation on performance of type-1 and interval type-2 Takagi-Sugeno fuzzy models

In the literature, there have been numerous studies demonstrating experimentally that type-2 fuzzy models outperform their type-1 counterparts. Although the advantages of these models seem to be well justified, the quantification of the improvements is not carefully evaluated and critically assessed in the existing studies. A thorough multi-objective experimental numeric evaluation of benefits of type-2 fuzzy models is still lacking. In this study, a numeric evaluation of the performance of type-1 and type-2 fuzzy models is carried out in terms of the criteria of accuracy and computing overhead, which leads to a thorough analysis of existing trade-offs between these two performance indexes. In the proposed numeric evaluation, type-2 fuzzy models are evaluated against their associated type-1 counterparts (the type-2 associated type-1 models sharing similar structure and the same development method). Three architectures of fuzzy models are involved in the comparative studies presented here: (1) fuzzy clustering method-based Takagi-Sugeno (TS) fuzzy models (Fuzzy C-Means based type-1, Fuzzy C-Means based interval type-2); (2) static TS-based fuzzy models (static type-1, A2C0, A2C1, EKFT2 and their associated type-1 models) and (3) evolving TS fuzzy models (SEIT2 and its associated type-1 counterpart, SCIT2 and its associated type-1 model). The experiments are carried out by involving 15 publicly available datasets. The accuracy of these two types of fuzzy models is assessed vis-a-vis their development time. Testing is involved to evaluate whether there are statistically significant differences between the performance of the type-2 and type-1 fuzzy models.