In a Decision Support System (DSS), data are used to make decision. When data are large, the accuracy of prediction is high. However, in many cases, data on hand is not enough but the decision has to be made. It causes a small data set learning problem. The way to solve this kind of problem is to improve forecast accuracy by current data. Adding some approximate data for granular computing can improve the forecast accuracy. In this article, Mega-fuzzification method that is based on Neuro-fuzzy method is first applied to add artificial data to improve learning accuracy. Considering the data bias phenomenon that often occurs in small data sets, this study provides a method that is based on the computational learning and probably approximately correct (PAC) theories for its adjustment as well as determines the assessment of the data domain. In addition, rough set method is used to help for Mega-fuzzification method to solve the problem of large number of attribute may affect the learning efficiency of Mega-fuzzification learning.
展开▼
