A novel low-parameter and fast density peak clustering

Density Peak Clustering (DPC) has attracted significant attention due to its ability to effectively identify clusters of arbitrary shapes. However, research on parameter-free density peak clustering remains relatively limited. Most existing algorithms rely on multiple parameters, and their reasonable tuning poses challenges. Meanwhile, traditional algorithms tend to select cluster centers in dense data regions, easily overlooking potential cluster structures in sparse regions. Additionally, the time complexity of DPC is O(n²), leading to inefficiency when handling large-scale datasets. To address these issues, this paper proposes a novel low-parameter and fast density peak clustering algorithm (PFDPC). The algorithm employs an adaptive mechanism, requiring only simple fine-tuning of one parameter. By combining natural neighbors with universal gravitation, it redefines local density to enable reasonable density estimation for samples in sparse clusters. Furthermore, a K-d tree is introduced to accelerate neighbor search, reducing the overall time complexity of the algorithm to O(nlogn). Experiments on synthetic datasets demonstrate the effectiveness of our method.