A robust clustering procedure for fuzzy data

In this paper we propose a robust clustering method for handling LR-type fuzzy numbers. The proposed method based on similarity measures is not necessary to specify the cluster number and initials. Several numerical examples demonstrate the effectiveness of the proposed robust clustering method, especially robust to outliers, different cluster shapes and initial guess. We then apply this algorithm to three real data sets. These are Taiwanese tea, student data and patient blood pressure data sets. Because tea evaluation comes under an expert subjective judgment for Taiwanese tea, the quality levels are ambiguity and imprecision inherent to human perception. Thus, LR-type fuzzy numbers are used to describe these quality levels. The proposed robust clustering method successfully establishes a performance evaluation system to help consumers better understand and choose Taiwanese tea. Similarly, LR-type fuzzy numbers are also used to describe data types for student and patient blood pressure data. The proposed method actually presents good clustering results for these real data sets.