K-means Algorithm for Optimizing Initial Clustering Center Based on Density Peak

The K-means is sensitive to the initial choice of cluster centers, leading to the results to be different every time. To address this, a new K-means variant based on decision values is introduced. This approach incorporates the nearest neighbor concept when calculating decision values to ensure density correlation and improve clustering stability. Additionally, a revised method for local density calculation is proposed. With the predefined number of clusters, data points with higher decision values are automatically selected as centers, and the median of points in each cluster replaces the mean to reduce the influence of outliers. Experimental results show that this decision-based K-means algorithm (DPK) achieves stable performance, requires fewer iterations, demonstrates robust handling of outliers, and offers improved clustering accuracy.