Machine Learning Methods for Multi-criteria Decision Making

点击翻译按钮获取中文摘要

Multi-criteria decision making remains an active and challenging research topic. The proposed research program aims to utilize machine learning methods, such as rough sets and granular computing, in rule induction for complex decision making. In particular, I will examine three-way decision making, especially when multi-criteria and multi-agent are involved. I will first study issues related to three-way decision making. With rough sets, rules are extracted and classified into three categories called the positive, negative and the boundary rules. However, in many cases, we may not be able to make necessary and timely decisions because there are too many uncertain decision rules. In addition, we should consider both the accuracy of prediction rules and the consequences of applying these rules in real applications. A possible solution is to lower expectations for accuracy by converting uncertain decision rules to certain decision rules. As accuracy is sacrificed, unacceptable consequences may occur in some cases. Game theory will be used to resolve such dilemmas by locating balanced positions that meet the needs of generalization and accuracy. Mechanisms to find equilibriums or thresholds will be examined. I will then examine real applications where multiple agents or multiple measures are involved in decision making. In these situations, multiple positive, negative and boundary regions result from using three-way decision methods. A balanced or consensus decision needs to be made. Traditionally, such decisions were made using majority, committee, or unanimous decisions. A combination of granular computing and game theory will be used to reach an intelligent consensus. Possible applications of this method are feature selection for text categorization and Web-based decision support systems. In summary, I will study the possibility of building hybrid intelligent systems that assist humans to make informative and wise decisions on complex problems involving multiple criteria and multiple agents. It is expected that this will broaden our knowledge of decision support mechanisms.