Shared Neighbors and Representative Point Assignment based Spatial-Temporal Density Peaks Clustering

In spatial-temporal clustering by fast search and find of density peaks, the definition of local density often fails to preserve sufficient neighborhood information, which can lead to the erroneous selection of cluster centers. Furthermore, sample assignment frequently depends solely on the spatial distance between cluster centers, resulting in inadequate spatial-temporal constraints. This often leads to the misclassification of samples that are spatially proximate but exhibit significant differences in temporal characteristics. To address these limitations, this paper proposes a shared neighbors and representative point assignment based spatial-temporal density peaks clustering (SNRP-STDPC) algorithm. By adopting a spatial-temporal weighted Euclidean distance as a unified metric and designing a similarity measure based on spatial-temporal Shared Nearest Neighbors, the algorithm defines a weighted shared nearest neighbor local density to improve the accuracy of cluster center selection. A representative point assignment strategy is designed to identify and lock uncertain samples. By combining these points with shared nearest neighbor similarity. the algorithm achieves precise error correction, ensuring assignment accuracy even under complex distributions. Experiments on synthetic datasets demonstrate that the proposed algorithm exhibits significant advantages across multiple evaluation metrics. Furthermore, experiments on a real-world earthquake dataset show that the algorithm can effectively identify and distinguish between foreshock and aftershock events.