Vysochyna O. Methods and models for classification and prediction of computer information systems states based on modified probabilistic neural networks

The thesis is dedicated to improvement of computer information network monitoring system by means of the development of methods and models for classification and prediction states of the elements and the computer information network. The most known systems of computer information network monitoring are examined, their comparative analysis is presented, general requirements are formulated and general architecture of the similar systems is synthesized. The offered solutions are not able to predict the computer information network state. Therefore in such systems it is necessary to include the additional modules of statistical data processing. Efficiency of different classification algorithms is estimated at recognition of the computer information network state. The analysis of classification algorithms was carried out by means of the Weka data analysis system. The analysis of classification algorithms showed that the most effective method of classification is neural network. The main quality scores of the computer information net-work that are fundamental in the providing services were analyzed. Depending on the accepted values of quality scores classes of service were selected. The requirements analysis of different traffic types encountered in the computer information network to net-work resources was carried out. Computer information network states were formalized and classified. A method for classification and prediction of computer information network states based on modified probabilistic neural networks is developed. To solve the prob-lem of prediction created an additional layer. Prediction is based on regression analysis. A method for calculating the value of the ac-tivation function width parameter and the method for calculating the training set are developed. On the basis of the method model for classification and prediction of router states and model for classification and prediction of computer information systems states are developed.