Artificial neural network; histogram; fuzzification; predictions of sales

This article describes a new approach which uses a histogram to fuzzify variables. We used a linguistic expression to form a training set output vector. The whole fuzzification process of the training set output vector is described in detail. This proposed method was verified on a real data set. We found out that the adaptation of a neural network by fuzzified output vectors has a considerably lower prediction error rate compared with another one without such transformation. Another advantage of the fuzzification approach is that only one neural network can be used for more various data sets with a high range of data attributes (units, thousands, millions). The proposed improvements increase the performance of neural networks, which is presented in the final part.