KNOWLEDGE EXTRACTION BASED ON DECISION TREES AND STOCHASTIC SEARCH

The problem of mathematical support development is solved to automate the extraction knowledge as production rules from the training data samples. The object of study is the process of constructing models of non-destructive quality control. The subject of study are methods of production rules extraction for synthesis of quality control models. The purpose of the work is to improve the efficiency of the process of production rules extraction for constructing models of quality control based on training samples. The stochastic method for the decision trees synthesis is proposed, which uses information about the informativeness of features, the complexity of the synthesized tree, as well as the accuracy of its recognition, which allows to form on the initial stage a set of tree structures, characterized by a simple hierarchy and low error recognition, in the process of search to create a new set of solutions with taking into account information about the significance of the features and interpretability of generated trees, which, in turn, provides the possibility of constructing a decision tree with a small number of elements (nodes and branches between them), and an acceptable recognition accuracy and retrieval based on it the most valuable instances. The software implementing proposed method is developed. The experiments to study the properties of the proposed method are conducted. The experimental results allow to recommend the proposed method for use in practice.