EFNN-Gen — a uni-nullneuron-based evolving fuzzy neural network with generalist rules

The evolving fuzzy neural networks have a high degree of interpretability and a high capacity for solving pattern classification problems. However, their accuracy could deteriorate when there are few samples for particular classes available, e.g., when new classes appear in the data stream. One way to improve these models’ performance is to include a priori knowledge in their data-driven architectural structure. This article proposes the new concept of generalist rules based on assessing the specificity of Gaussian functions that make up the neurons of the first layer of the evolving fuzzy neural network (EFNN-Gen). These rules can be seen as general (expert) knowledge about the classification problem. Tests performed with real binary pattern classification bases proved that integrating generalist rules can increase the accuracy of an evolving system and that there is a specific limit on how many rules can be used for this improvement.