The Research Of The Model Selection In Decision Support System

With the development of computer science and technology, computer is applied to all of life. The model system of Decision Support System is also more and more attention to people.Early the model system of Decision Support System storages the model as a data or a subroutine, and manages models by a method of similar data management. With the increasing complexity of decision-making and needs of support for knowledge processing, this simple model system can not meet the new decision-making needs.These years, in the aspects of model selection method and model selection knowledge management has made some progress. However, since model selection is a difficult problem, so far the results in this field are still limited, the vast majority of research results are also not achieve automatic model selection, but rather depend on experts and past choice experience to choose. Based on this scenario, the automatic model selection has a certain theoretical value and far-reaching significance.This paper, after study of decision tree classification algorithm, direct towards the rule generation methods, improve decision tree algorithm. Add a coefficient defined based on expert's priori and domain knowledge to calculation formulas of some property information entropies and information gains. Transform weighted sum into sum of weighted sum and weighted coefficients, which can increase some attribute information entropies, and corresponding reduction other property information entropies. The decision tree established by the improved decision tree algorithm is more compact and simple than the original, and it can not only speed up the growth of decision tree but also can get a better structure decision tree, in order to excavate better rules from information.End, this paper has a case study in the decision-making model established by the improved decision tree algorithm by forming a decision-making of a supermarket. And has a analog simulation test for the decision-making model using four data sets of UCI public database. Experiments show that the improved algorithm established a more concise decision-making model based on does not change the prediction accuracy of the original algorithm basically. Reduce decision makers'labor input and it is more easily used by decision makers. At last, conclude the results of this institute and propose the improved part and the shortage for study in the future.