| 摘要 |
<p>Clustering the data is challenging and difficult task with respect to optimization because most of the existing practical clustering algorithms like k-means, k-modes, gradient descent and so on, converges to local optimum or terminates at local optimum. These methods are based on the selection of initial cluster centroid and are extremely sensitive to these initial points. So, by definition, it is sensitive to initial points. In most of the existing clustering methods, selection of this initial point was made randomly. The points in particular cluster depend on the properties of this random point. Hence different initial points lead to different clustering results. Instead of this random selection, here our invention introduces a hybrid approach to select initial cluster centroid for various clustering applications and demonstrated with popular k-modes algorithm. This invention will act as an initializer/ preprocessor for any clustering algorithm for which centroid plays a key role, by refining the initial point prior to clustering. Following invention is described in detail with the help of Figure 1 of sheet 1 showing the steps of the proposed method, Figure 2 of sheet 2 showing the extension of step 5 of figure 1and Figure 3 of sheet 3 showing generation of initial cluster centroid using proposed method.</p> |
