Growing fuzzy inference neural network applied to the NN3 neural forecasting competition

Growing fuzzy inference neural system (GFINN) is a fuzzy–neural network model. Its functionality can be expressed in a form of fuzzy if–then rules. The skill of the GFINN model to grow allows it to change its size and structure according to the training data. The resulting structure allows for a simple input features selection — not all input features have to be used in every fuzzy rule. The new algorithm for the computation of output weights runs much faster than least mean squares estimate, while the experiments performed in past show nearly identical performace of both methods. Furthermore, the new training method guarantees the output weights to remain inside the output values interval. This makes the extracted rules reasonable, unlike with least mean squares estimate. This text describes the GFINN model and its application to the NN3 neural forecasting competition.