A Domain Adaptive Density Clustering Algorithm for Data With Varying Density Distribution

As one type of efficient unsupervised learning methods, clustering algorithms have been widely used in data mining and knowledge discovery with noticeable advantages. However, based on density peak limited effect varying distribution (VDD), equilibrium (ED), multiple domain-density maximums (MDDM), leading to the problems sparse cluster loss fragmentation. To address these problems, we propose a Domain-Adaptive Density Clustering (DADC) algorithm, which consists three steps: domain-adaptive measurement, center self-identification, self-ensemble. For VDD features, clusters regions are often neglected by using uniform thresholds, results clusters. We define measurement method K K-Nearest Neighbors (KNN) adaptively detect peaks different regions. treat each point its KNN neighborhood as subgroup better reflect domain view. In addition, for ED or MDDM large number similar values can be identified, self-identification self-ensemble automatically extract initial centers merge fragmented Experimental demonstrate that compared other comparative algorithms, proposed DADC algorithm obtain more reasonable VDD, features. Benefitting from few parameter requirement non-iterative nature, achieves low computational complexity is suitable large-scale clustering.