machine learning - More accurate approach than k-mean clustering -

in radial basis function network (rbf network), prototypes (center vectors of rbf functions) in hidden layer chosen. step can performed in several ways:

• centers can randomly sampled set of examples.
• or, can determined using k-mean clustering.

one of approaches making intelligent selection of prototypes perform k-mean clustering on our training set , use cluster centers prototypes. know k-mean clustering caracterized simplicity (it fast) not accurate.

that why know other approach can more accurate k-mean clustering?

any appreciated.

several k-means variations exist: k-medians, partitioning around medoids, fuzzy c-means clustering, gaussian mixture models trained expectation-maximization algorithm, k-means++, etc.

i use pam (partitioning around medoid) in order more accurate when dataset contain "outliers" (noise value different others values) , don't want centers influenced data. in case of pam center called medoid.