DBSCAN-based granular descriptors for rule-based modeling

Rule-based modeling is a useful approach in modeling both complex nonlinear and non-numeric systems, e.g. having linguistic information. However, modeling complex systems in big data era brings new challenges for conventional rule-based modeling, such as high computation overhead and low representation ability of rule. To address these problems, this paper proposed an advanced rule-based modeling method-based DBSCAN-inspired granular descriptors. First, to understand the essential characteristics of data and enhance rules' representation ability, data structures are obtained by DBSCAN clustering algorithm, which has high flexibility at coping with diverse geometry. Second, numerous granular descriptors are constructed in the refined representation of data structures and used for fuzzy rule formation. This granular computing process could effectively reduce computation overhead of big data analysis. Finally, the proposed rule-based model consists of fuzzy rules and interval outputs, which are resulted from structural granular descriptors and justifiable granulating respectively. Experimental studies concerning synthetic data and publicly available data illustrated that the proposed method can achieve prior performance on both modeling and time consuming than conventional rule-based modeling via FCM. Therefore, it is verified the developed approach is feasible and useful to be applied in modeling real complex engineering systems.

Automated summary

This paper proposes an advanced rule-based modeling method based on DBSCAN-inspired granular descriptors to model complex nonlinear and non-numeric systems. The method enhances the representation ability of rules by obtaining data structures through the DBSCAN clustering algorithm and constructing granular descriptors. Experimental results demonstrate that the proposed method outperforms conventional rule-based modeling method FCM in both modeling and time consumption.