Research On Optimization Of Combination Weighting Based On Item Discrimination

Multi-attribute decision making has been widely applied in many areas. Many kinds of multi-attribute decision making method are proposed. In many of these methods, the effect that attribute weight has on final decision is very sensitive. Evaluating alternative based on attribute weight, alternative value is found very close to each other, especially the alternative which are ranked very high. So it is not easy to make decision. In order to improve decision performance, it is necessary to do research on the relation between combination weighting and item discrimination.The paper follows the idea bellow:Firstly, previous research is reviewed and the framework of the paper is pointed out. Then, item discrimination is definited, and the effect that dimensionless of data and alternative quantity have on item discrimination is analyzed. Next, two phase optimization algorithm of combination weighting is proposed. In the first phase, the paper selects some alternatives by statistical method, to reduce several alternatives. In the second phase, a combination weighting model is supposed, combining two goals: minimizing the deviation between combination weight and subjective weight and objective weight; maximizing the deviation between alternatives. Meanwhile, modification algorithm of combination weight based on item discrimination is proposed. In modification, criteria for judging combination weight is designed, and the steps of weight modification are proposed according to statistical methods. Finally empirical research is done.The contribution of the paper is illustrated as below:1) The definition of item discrimination is proposed, and two kinds of factors that may have influence on item discrimination have been analyzed.2) Based on the two phase optimization, a new algorithm of combination weighting is proposed, from the aspect of alternative selecting and optimization of combination weighting.3) To solve the problem that item discrimination is still not significant after optimization of combination weighting, modification algorithm of combination weight based on item discrimination is proposed.