Training set fuzzification towards prediction improvement

Abstract

This article presents a method of fuzzification of variables using a histogram. This approach is used when creating an output vector of a training set that forms linguistic variables. An appropriate transformation of an input vector of the training sets was also proposed. Both of the aforesaid procedures were described in detail in the article. An extensive comparative experimental study with the following outcomes was carried out. The neural net which was adapted by the transformed training set showed a significantly better prediction than a neural network which was adapted by a training set without making any changes. The results of this experimental study were analyzed in the conclusion.