A Novel Improved and Synthesized Evaluation Model for Multi-Metrics
Shen-Bo Xu,Yun-Cheng Man,Wen-Kun Dou,Zhong-Hao Wu,Lin-Chuan Zhao
DOI: https://doi.org/10.1142/9789814759687_0012
2016-01-01
Abstract:Assume that there are p samples and q metrics, which also means the data has g dimensions. For those original accurate data, we employ the normalization method to transfer them into numbers ranging from -1 to +1; for those indicators which cannot be presented as a number, we apply the man-made scoring: -1, -0.5, 0, +0.5, + 1in five levels. Then separately, we performed Analytic Hierarchy Process and Maximum Entropy Model on these secondary data. By multiplying the weight and the correlation coefficient, Level 3 data is obtained. Finally, the Principal Component Analysis is carried out, and we can derive the function between the original data and the evaluation result. However, the connection is quite unstable if any data changes significantly. So we put forward a feedback system, which evaluates the final result (as a number) to be a new original data. Apparently, this model still has some contingencies. In order to evaluate the error, we establish an implicit relationship between the original data and the final results by Artificial Neural Networks Principals. Compared the simulation without previous results, one may conclude that our model is robust and valid under most circumstances. Finally, we can work out the error percentage of our evaluation model.