An interpretable evolving fuzzy neural network based on self-organized direction-aware data partitioning and fuzzy logic neurons

This paper proposes the definition of architecture an evolving fuzzy neural network based on self-organizing direction aware data partitioning through stochastic processes dataset used in model. The choice between four different types logical neurons second layer and six neuron activation function that compose artificial aggregation are defined a procedure generating random pairs combinations two factors. combination obtains maximization training accuracy is chosen to perform composition model structure, whose components can be transferred readable IF-THEN rules for interpretability purposes. Furthermore, able adapt its parameters evolve structure autonomously with new samples self-organized direction-aware (SODA). In this context, we also propose technique measure degree changes (rules), which could structural active learning purposes (e.g, request user feedback case significant changes). To compare proposed approach, binary pattern classification tests were performed, results compared other models networks networks, obtaining satisfactory elements their comparing final when classifying real datasets. obtained best result 3 synthetic bases evaluated, addition patterns five nine evaluated It highlights problems patients who underwent breast cancer surgery (72.44%), diabetes evaluation (67.87%), Australian (72.27%) German (80.51%) credit ratings evaluation, finally radar signals ionosphere (90.46%). noteworthy obtain evolution extracted from problem identification respiratory diseases collection saliva 76.67% accuracy. presented by superior traditional intelligence models, while it was possible extract knowledge form interpretable realize how these changed over time.