Distance-weighted K-means Clustering Algorithm Based on Local Density
Zhuoheng Lv, Qingxin Liu · 2023
The traditional K-means clustering method is often affected by the initial selection of center points, as well as the presence of outliers and edge points during the iterative process. To address these limitations, this paper proposes a distance weighted K-means algorithm based on local density, known as the LDDW-K-means algorithm. This algorithm utilizes local variance to assess the density of the region where the sample points are located. It also incorporates an improved roulette wheel method to heuristically select high-density sample points that are widely separated as the initial clustering centers. During the iterative process, an adjusted cluster radius is employed to mitigate the impact of outlier and edge points on the cluster centers. Additionally, the sample points within the adjusted cluster radius are weighted based on their distances to obtain the new cluster center coordinates. Experimental results on the UCI dataset demonstrate that the LDDW- K-means algorithm effectively enhances clustering performance metrics such as accuracy (ACC) and adjusted mutual information (AMI).