A generalized framework for ANFIS synthesis procedures by clustering techniques

The application of machine learning and soft computing techniques for function approximation is a widely explored topic in literature. Neural networks, evolutionary algorithms and support vector machines proved to be very effective, although these models suffer from very low level of interpretability by human operators. Conversely, Adaptive Neuro Fuzzy Inference Systems (ANFISs) demonstrated to be very accurate models featured by a considerable degree of interpretability. In this paper, a general framework for ANFIS training by clustering is proposed and investigated. In particular, different derivative-free ANFIS synthesis procedures are considered for performance evaluation, by taking into account different clustering algorithms, dissimilarity measures and by including an additional neuro-fuzzy classifier downstream the clustering phase targeted to rule base refinement. The resulting ANFISs have been compared, in terms of effectiveness and efficiency, on several benchmark datasets against three suitable competitors, namely a Support Vector Regression, MultiLayer Perceptron and a K-Nearest Neighbour decision rule. Computational results show that the proposed techniques tend to outperform competing strategies while, at same time, featuring models with lower structural complexity. A complete software suite implementing the proposed framework is freely available under an open-source licence.