Granular neural networks computing on fuzzy information table

Abstract

Granular computing is the key to granular neural networks, and in fact it is also the main problem in knowledge discovery and data mining. This paper addresses fuzzy information extraction and granular computing in granular neural networks in order that fuzzy rules can be discovered from fuzzy information which is difficult to be measured accurately with numerical data and furthermore the missing data can be predicted. An information table is modeled and the relation embedded in data has been discovered through granulation. A fuzzy neural network model is proposed to learn with a given fuzzy information table in crisp granular neural network and fuzzy granular neural network respectively, then to predict missing rules in a larger scale information table. The conclusion based on experiments is that granular neural networks can be used in knowledge discovery embedded in fuzzy data base and granular computing is accomplished. The developed techniques have promising applications in stock-markets forecasting especially expressing not only technical factors but also fundamental aspects which are hard to be quantified.