Abstract:
In cluster analysis, the selection of the initial central point plays a crucial role on the results of the cluster. By using the sample density of information, this paper proposes an algorithm to obtain the central point. This algorithm can obtain the central point without setting any parameters. And then, by training the obtained central point via competition network, the central point can be made closer to the center of each cluster. The simulation results show that this algorithm is effective and high reliability, and can ensure that the central point to be located in a different type of cluster before the training of competition network, which can improve the efficiency of the network.