I'm using WEKA- Simple K-means clustering algorithm in order to get the efficient and accurate pattern for my unsupervised training data. My training data is a medical (diabetes) data that consists of 940 instances, and 20 numerical and nominal attributes represent the values of home and lab test readings, disease risk diagnosis scale (scale is from 1-10) and 8 recommended treatments ( ranked 0 or 1). I read somewhere that one of the methods that improve the performance of kmeans clustering is to filter the data through attribute selection. And as I'm using WEKA, I've tried using different attribute selection algorithms which have ranked my attributes, and then run clustering. But nothing changed and no improvement seen since the WSS (within cluster sum of square error) is the same before and after filtering. Anyone can help and advice me in that? and what are the other methods in WEKA that help in enhancing the clustering performance specifically simple k-means.
Thank you, Asmaa