Intelligent Data Analysis

ALL has developed a toolkit to realise data analysis, evaluation of data and of the results of analysis. These functions can be done within the context of the models and within their dependence on the problems to be solved. The methodology provides various methods of analysis such as logical, statistical and fuzzy depending on the needs of the problem to be solved.

The methods of intelligent data analyses can be grouped as follows:

Methods proposed for the analysis: special logical methods, learning methods for pattern recognition and method for interactive definition of solution and decision rules for pattern recognition;

The logical intelligent data analysis method generates hypotheses about the cause-effect relationships and provides the best explanation by selecting the right one from the set of hypotheses.

Original methods are proposed for the extraction of informative components in order to construct simplified qualitative models on them, and so find regularities in the data base. These methods of data analysis also use learning and modelling;

Methods for model building for hardly formalizable fields augmented by tools which have been specially developed to consider data of poor quality, or noisy and incomplete data. Here data can be textual, numerical and graphical;

The main features of the methods that have been developed are as follows:

• flexibility, that permits to work with data of poor organization and of different difficulty;
• samples with small number of elements with description of big size (e. g. quantity of properties can be in the order of a few hundred thousands while the number of observations is a few hundred), can be processed
• samples of a description consisting of a few hundred thousand elements with a few hundred quantitative features can also processed,
• visualisation, visual support of the analysis to support the realization of cognitive computer graphics;
• extraction and utilization of expert knowledge
• utilization of directed threshold nets
• to represent the expert knowledge
• for data processing and data output in any interpretable form;
• application of automatic learning of pattern recognition and modeling by neuron-like decision schemes.

An original approach was developed to support modelling, intelligent data analysis and forecasting. The approach provides efficient methods

•  to create constructive models (they support efficient experimentation),
• to combine different types of models such as static and dynamic, logical and numerical, statistical and deterministical ones

In the approach self-organising models play a central role.

The methods of the proposed approach are as follows:

• Three classes of methods for self-organising modelling
• self-organizing group method of data handling,
• bistable gradient neural network,
• threshold network.
• Special statistical methods
• Special logic-based method for supporting descriptive (qualitative) information.