An integrated approach for the identification of compact, interpretable and accurate fuzzy rule-based classifiers from data

This paper presents three very simple and computationally undemanding symbiotic algorithms for the identification of compact fuzzy rule-based classifiers from data. The problem of interpretability is specifically addressed, resulting in a conclusion that due to the characteristics of classification tasks a major well-known interpretability condition - distinguishability - can be discarded. It is shown that despite the interpretability-accuracy tradeoff, accuracy of identified classifiers stands out to comparison. All obtained properties can be very useful in practical problems. The proposed method is validated on Iris, Wine and Wisconsin Breast Cancer data sets.