TWRBF - Transductive RBF Neural Network with Weighted Data Normalization

This paper introduces a novel RBF model – Transductive Radial Basis Function Neural Network with Weighted Data Normalization (TWRBF). In transductive systems a local model is developed for every new input vector, based on some closest to this vector data from the training data set. The Weighted Data Normalization method (WDN) optimizes the data normalization range individually for each input variable of the system. A gradient descent algorithm is used for training the TWRBF model. The TWRBF is illustrated on two case study prediction/identification problems. The first one is a prediction problem of the Mackey-Glass time series and the second one is a real medical decision support problem of estimating the level of renal functions in patients. The proposed TWRBF method not only gives a good accuracy for an individual, “personalized” model, but depicts the most significant variables for this model.