Adaptive Fuzzy Neural Trees

We propose Adaptive Fuzzy Neural Trees as an appropriate tool for intelligent data analysis, comprehension , and prediction. Instead of using a single technique Adaptive Fuzzy Neural Trees as a mixture of paradigms combine the main advantages of neural networks, decision trees, and fuzzy logic. Like neural networks they are able to model smooth functions and can be adapted incrementally. Like decision trees their topology and initial parameters can easily be derived from a training data set by well known statistical methods, they provide for an extremely eecient implementation, and they allow for insights to human experts. As with fuzzy systems human knowledge can be brought in and soft decisions can be made with respect to fuzzy sets. Our experiments with an implementation of Adaptive Fuzzy Neural Trees shows that the multiple paradigm concept has clear advantages over the pure single paradigm approaches.