Informatica

This research work is aimed at the development of data analysis strategy in a complex, multidimensional, and dynamic domain. Our universe of discourse is concerned with the data mining techniques of data warehouses revealing the importance of multivariate structures of social‐economic data which influence criminality. Distinct tasks require different data structures and various data mining exercises in data warehouses. The proposed problem solution strategy allows choosing an appropriate method in recognition processes. The ensembles of diverse and accurate classifiers are constructed on the base of multidimensional classification and clusterisation methods. Factor analysis is introduced into data mining process for revealing influencing impacts of factors. The temporal nature and multidimensionality of the target object is revealed in dynamic model using multidimension regression estimates. The paper describes the strategy of integrating the methods of multiple statistical analysis in cases, where a great set of variables is observed in short time period. The demonstration of the data analysis strategy is performed using real social and economic development of data warehouses in different regions of Lithuania.

The exact solution of the reliability of structures under stochastic loading is generally difficult, and various approximate methods have been developed. The most popular are the linearization method, the Monte-Carlo method and its numerous variants. In this paper new modification of the Monte-Carlo method based on asymptotical expansion is examined. Results of mathematical simulation are given.

In the paper a general approach to identification of non-linear autoregression processes in the class of parametric and non-parametric mathematical models is formulated. With the help of mathematical simulation the estimates of the processes of this class are studied: a nuclear estimate, an estimate of least squares projective estimates. Some statistical properties of these estimates are indicated.