A Comparative Analysis of Fuzzy Clustering and K-means Clustering for 15-day Rainfall Datain DIY Province.

Abstract

This study aimed to compare performance two clustering methods (fuzzy c-means and K-means) on the basis validity index. The study used daily rainfall data for twenty five years (1985 to 2009) from 22 stations covering the DIY Province. PCA is used to reduce number 15-day rainfall variables and transform into new variable. Four validity index clustering: Xie-Beni index (XB), sum squared error (SSE), silhouette (Si) and standard deviation ratio (S w /S b ) are used for compare performance of two clustering methods. The optimal number of cluster is determined using XBindex, and result SSE, Si and ratio (S w /S b ) are compared to find appropriate clustering algoritms to 15-day rainfall data. The study results showed that the first 4 PCs explains more than 82 % of total variance, than are used for data input in fuzzy c-means and K-means algorithms. The optimal number clusters according XB index are sixand five forK-means and FCM methode, respectively.The average ratio S w /S b K-means methode (0.243) was smaller than fuzzy c-means methode (0.289). The Silhouette dan SSE index are 0.46 and 76, 0.24 and 254 for K-means and FCM respectively. The result indicated that K-means methodeto be better than fuzzy c-means for clustering 15-day rainfall data in DIY Province.