Improved Fuzzy Kernel Clustering With Outliers

Outliers are data values that lie away from the general clusters of other data values. It may be that an outlier implies the most important feature of a dataset. Their identification is important not only for improving the analysis but also for indicating anomalies which may require further investigation. In this paper, a comparison of initializing prototype V and membership degree U is made, which shows that initializing V has many advantages over those available. We introduce a new objective function for fuzzy clustering, which uses V kernel distance as the function, and the derivation procedure of the function is also shown in the paper. A comparison between the paper's and the available method shows that the new objective function has the advantages of less search space. It is also shown that our method has less time complexity than the available. The simulations demonstrate the feasibility and speedy of the proposed method. Firstly, experiments of different datasets by applying different kernel function and objective function are made. Secondly, experiments of initializing different V are made. Finally, figures of membership degree U and outlier weighting factor W are shown. Our method can greatly increase the convergence speed and decrease the algorithm's processing time under the good clustering result is kept.