An Entropy-Based Uncertainty Measure for Developing Granular Models

There are two main ways to construct Fuzzy Logic rule-based models: using expert knowledge and using data mining methods. One of the most important aspects of Granular Computing (GrC) is to discover and extract knowledge from raw data in the form of information granules. The knowledge gained from the GrC, the information granules, can be used in constructing the linguistic rule-bases of a Fuzzy-Logic based system. Algorithms for iterative data granulation in the literature, so far, do not account for data uncertainty during the granulation process. In this paper, the uncertainty during the data granulation process is captured using the fundamental concept in information theory, entropy. In the proposed GrC algorithm, data granules are defined as information objects, hence the entropy measure being used in this research work is to capture the uncertainty in the data vectors resulting from the merging of the information granules. The entropy-based uncertainty measure is used to guide the iterative granulation process, hence promoting the formation of new granules with less uncertainty. The enhanced information granules are then being translated into a Fuzzy Logic inference system. The effectiveness of the proposed approach is demonstrated using established datasets.