A modified kernel clustering method with multiple factors

We propose a simple and effective method about kernel clustering. This method takes many factors about kernel clustering into account. These factors include the selection of the initial centers of kernels, the ways of how to compute widths of kernels and the distances between patterns, different growing ways of kernels, and different kernel clustering criterions. Experiments have validated that these factors have influence on the final experimental results while not each factor has a great influence. Furthermore, some classifiers with this proposed kernel clustering method have higher classification accuracies and lower generalization risks.