Methods for Summarizing Multi-Criteria Decision Table(3/3)

In our daily life, we always face with various situations that need to make decision.Accordingly, various approaches have been proposed and applied to help making decision,such as operation research, statistical analysis, mathematics, machine learning, fuzzy theory,and so on. One of the most important approaches is Multi-criteria decision making (MCDM)model. Generally, a typical form of MCDM may include three parts, i.e., input, output andsolution approach. The input can be expressed as a decision table of m alternatives with ncriteria, where the value of each cell reflects the performance of an alternative on a certaincriterion. In addition, the output result is typically a single optimal solution or a set ofefficient solutions preferred by decision maker. Although MCDM can make decision onbehalf of decision makers by directly generating final alternatives, it did not pay enoughattention on how to provide critical information to support decision makers to analyze theirown problems.In light of this weakness, we propose a method which can summarize decision table andrepresent it as a simplification table for decision maker to support their making decision. Thetable is called information summarization table, which is easier to use and understand andanalyze for decision makers. Furthermore, the summarization table can be represented byradar chart to display how well each group of alternatives performs on each group of criteria.For the problem above, three research issues are defined in this project. In the first year, wewill design a GA algorithm to obtain an st summarized table from the original mn decisiontable, and the information lost degree resulted in the summarization process should be assmall as possible. In the second year, a greedy-based heuristic algorithm with incrementalcapability to compute objective value will be proposed to obtain an st summarized tablefrom the original mn decision table. In this heuristic algorithm, the values of s and t aredetermined automatically by the algorithm instead of specified by users before the executionof the algorithm. Finally, in the last year we will propose a heuristic algorithm with randomsearch capability to obtain an st summarized table from the original mn decision table. Theaim of including random search capability into our heuristic algorithm is to expand thesolution search space and jump out of local optimal trap caused by the greedy search strategyused in the seond algorithm.