MULTILAYER ADAPTIVE FUZZY PROBABILISTIC NEURAL NETWORK IN CLASSIFICATION PROBLEMS OF TEXT DOCUMENTS
Yevgeniy V. Bodyanskiy, Nataliya Ryabova, Oleh Zolotukhin · Radio Electronics Computer Science Control · 2014
The problem of text documents classification based on fuzzy probabilistic neural network in real time mode is considered. A differentnumber of classes, which may include such documents, can be allocated in an array of text documents. It is assumed that the data classes canhave an n-dimensional space of different shape and mutually overlap. The architecture of the multlayer adaptive fuzzy probabilistic neuralnetwork, which allow to solve the problem of classification in sequential mode as new data become available, is.proposed. An algorithm fortraining the multilayer adaptive fuzzy probabilistic neural network is proposed, and the problem of classification is solved on the basis of theproposed architecture in terms of intersecting classes, which allows to determine the belonging a single instance of a text document to differentclasses with varying degrees of probability. Classifying neural network architecture characterized by simple numerical implementation and highspeed training, and is designed to handle large data sets, characterized by the feature vectors of high dimension. The proposed neural networkand its learning method designed to work in conditions of overlapping classes, differing both the form and size.