Application of neuro-fuzzy ensembles across domains: A systematic review of the two last decades (2000–2022).

Neuro-fuzzy systems have received considerable attention from academia in the last decade. They can provide a tradeoff between the performance of artificial neural networks and the interpretability of fuzzy inference systems expressed through fuzzy rules. Single techniques do not always provide the optimal performance, especially when the results are susceptible to changes in the data or hyperparameters. Therefore, ensemble learning can be used to build a more stable and robust model. In this paper, a systematic literature review of studies evaluating neuro-fuzzy ensembles was performed to highlight the importance of ensemble learning and its role in improving the performance of neuro-fuzzy systems. Many aspects are highlighted, including publication years, sources, contribution types, application domains, single neuro-fuzzy systems, ensemble techniques, and the overall performance and interpretability. A total of forty-eight articles published from 2000 to March 2022 addressing the use of neuro-fuzzy ensembles in different engineering applications were selected and analyzed. Results show that: (i) Takagi–Sugeno–Kang neuro-fuzzy systems are the most evaluated category, especially the ANFIS single. (ii) Homogeneous​ ensembles, in particular boosting, are the most investigated. (iii) Logical, relational and ANFIS ensembles are outperforming. And (iv) The interpretability of neuro-fuzzy ensembles was not thoroughly investigated, but based on the number of rules, logical systems were the best. [ABSTRACT FROM AUTHOR]