Method for determining comprehensive weight vector based on multiple linear fitting

Based on evaluation vectors derived from single factor and the comprehensive evaluation results of some known typical objects, the sample data is classfied into different kinds, and the weight vector of the multi-factor comprehensive evaluation problem is transformed into the solution of an overdetermined systems, which consist of inconsistent linear equation constituted by the vectors of single factor evaluation and the comprehensive evaluation results. The approximate numerical relationship of the standard sample data from the typical objects is fitted with the least squares method, so that the weight vectors to different kinds of objects for comprehensive evaluation are obtained. A learning and correction way to find a multiple linear weighted synthesis model with minimum number of sub-model for the classification of object is presented. A typical object is employed as the center for every sub-model. In applications, the objects are assigned to that class whose center is closest to the objects firstly, and the evaluation result can be derived by the corresponding sub-model. Finally, the effectiveness of the proposed method is illustrated by an example.