A Study of Support Vector Regression-Based Fuzzy c-Means Algorithm on Incomplete Data Clustering

抄録

Support vector regression-based fuzzy c-means algorithm (SVR-FCM) clusters data according to their relationship among attributes, which can provide competitive clustering results for the dataset having functional relationship among attributes. In this paper, we study the performance of SVR-FCM on incomplete data clustering. The conventional incomplete data clustering strategies of fuzzy c-means algorithm (FCM) are first applied to SVR-FCM, and a new strategy named MIS strategy is designed to assist SVR-FCM handle incomplete data as well. A number of synthetic datasets are used to study the effect of data missing rate and missing attribute numbers on the performance of SVR-FCM based on different incomplete data clustering strategies. Several engineering datasets are used to test the performance of the current and proposed incomplete data clustering strategies for SVR-FCM. The results indicate that SVR-FCM can provide better clustering results than FCM for the dataset having functional relationship among attributes even if it has missing values, and the proposed MIS strategy can assist SVR-FCM to achieve the best clustering results for most datasets.