A K-prototypes Algorithm Based on Improved Initial Center Points

The K-prototypes is the main clustering algorithm that capable of handling mixed numeric and categorical data.However,Kprototypes sensitive to its initial center points,is parameter-dependent and susceptible to noise interference.In order to overcome them,a method is proposed to build initial center points heuristically through the neighbors of objects,and then calculate according the K-prototypes algorithm's procedures.At last,use a rule to optimize the clustering results which able to identify the abnormal points.The proposed algorithm successfully resolved the defects of the traditional algorithm,improves the accuracy of clustering results and stability of the algorithm.Experiments show the proposed algorithm leads to better accurate and scalable,superior to the traditional K-prototypes.