Aug 2026· Journal of King Saud University: Computer and Information Sciences· Vol 38· 0 citations· 32 references
TL;DR
Experiments show that MFGB-DBSCAN achieves competitive clustering accuracy and efficiency compared with representative baselines, particularly on datasets with varying densities and complex structures.
Abstract
Density-based clustering algorithms such as DBSCAN can identify clusters with arbitrary shapes, but their performance is often limited by global parameter sensitivity and high computational cost on complex datasets. Granular-ball-based methods improve efficiency by replacing point-level operations with coarse-grained hyperspherical units, yet existing GB-DBSCAN-style approaches still suffer from limited adaptivity, unreliable granular-ball fusion, and ambiguous boundary assignment. To address these issues, this paper proposes MFGB-DBSCAN, a multi-factor granular-ball-based density clustering algorithm with adaptive parameter optimization. The proposed method follows a reliability-aware coarse-to-fine strategy. First, adaptive parameter initialization estimates key neighborhood and connection scales from data distribution characteristics. Second, a quality-driven granular-ball evaluation mechanism integrates density, cohesion, and separation to distinguish reliable structural units from uncertain boundary regions. Third, a multi-factor fusion strategy combines spatial proximity, density consistency, quality alignment, geometric overlap, and cohesion compatibility to construct cluster skeletons. Finally, a confidence-based two-stage label propagation mechanism refines the assignment of boundary samples. Experiments on synthetic and real-world datasets show that MFGB-DBSCAN achieves competitive clustering accuracy and efficiency compared with representative baselines, particularly on datasets with varying densities and complex structures. The results suggest that MFGB-DBSCAN provides a practical and reliability-aware improvement for granular-ball-based density clustering.
Density Peak Clustering (DPC) has emerged as a powerful clustering algorithm capable of identifying clusters of arbitrary shapes by detecting density peaks in data. However, DPC suffers from several limitations: manual selection of cluster centers from the decision graph, sensitivity to the cutoff distance parameter, and suboptimal assignment of border points. We propose TGDPC (Topology-Guided Density Peak Clustering, Figure 1), a novel algorithm that address these limitations through an adaptive strategy combining graph connectivity analysis with density peak detection. TGDPC constructs a degree-constrained k-nearest neighbor graph to identify natural data partitions, employs delta-based center selection for more robust center identification, and implements an intelligent multi-stage merging strategy to guarantee exactly K clusters. Graph preprocessing is a natural approach to data preprocessing. Extensive experiments on synthetic and real-world datasets demonstrate that TGDPC achieves significantly superior clustering performance over classic algorithms (e.g., K-means, DBSCAN, and Spectral Clustering) as well as state-of-the-art DPC variants such as DPC-DBFN, DGDPC, WANN-DPC, and VDPC. The algorithm maintains O(n2) time complexity while providing deterministic, parameter-robust results without requiring manual intervention.
Heng Wu· International Conference on...· 0 citations
Granular-Ball Computing (GBC) is an efficient, robust, and highly interpretable multi-granularity representation and computation method. Nonetheless, most feature selection methods based on GBC require considerable time to calculate the significance measures of features or repeatedly generate granular balls, which limits their applicability to high-dimensional data. The graph-based feature selection effectively reduces dimensionality by exploiting feature correlations and redundancies. However, most graph-based feature selection methods are limited to a fine and single granularity knowledge space. Driven by these issues, this paper first proposes the granular-ball divergence-based fuzzy rough set to characterize the uncertain information from a multi-granularity perspective. Then, the minimum discriminative criterion for constructing a hypergraph is evaluated by the approximation operators, and the correlations between theories are established. On this basis, the importance of features is defined as the weights of hypernodes, and the strategies of Important Retaining (IR) and Redundant Pruning (RP) are designed to select the most important feature and improve execution efficiency, respectively, which is equivalent to an iterative weighted maximum coverage problem with dynamic weight updates. Finally, a feature selection algorithm is designed to select the best feature subset. The experimental results show that our algorithm achieves better classification performance and higher execution efficiency.
Ye Li, Lei Yang, Binbin Sang et al.· IEEE Transactions on Knowled...· 1 citation
Clustering is a fundamental data mining technique that groups data points by similarity. A critical challenge for clustering algorithms is the effective selection of initial cluster centers, often done through inefficient trial-and-error. To address this, a novel Adaptive Cluster Center Initialization using Density Peak for Geodesic Distance-based Clustering (AGDPC) method intelligently identifies optimal centers. It builds upon the Density Peaks Clustering (DPC) and Geodesic-Based Initialization (GDPC) approaches, using weighted Euclidean distances that incorporate the Pearson correlation coefficient. Furthermore, AGDPC adaptively optimizes its threshold parameter using data field density estimation entropy, enhancing its accuracy and automation. It should be noted that, AGDPC currently operates directly on high-dimensional data without dimensionality reduction. While this preserves the original structure of the data, high-dimensional spaces often suffer from the “curse of dimensionality,” where distances between points become less meaningful, and computational complexity increases. Introducing Multidimensional Scaling (MDS) as a pre-processing step can address these challenges by projecting data into a lower-dimensional space while preserving pairwise distances or dissimilarities as much as possible. In this work, we propose to use multidimensional scaling that mitigates the impact of irrelevant or redundant features, enabling 2D/3D visualisation of clusters for interpretability. We proposed to use Regularised Multidimensional Scaling using Radial Basis Function (RBF-MDS) for dimension reduction. The idea is to decrease its dimensionality, followed by the application of the adaptive strategy on the dataset in its reduced dimension. Our approaches are tested on various benchmarking datasets, and the outcomes are contrasted with those of DPC, GDPC, and one of the most conventional clustering methods, K-Means clustering. When tested on actual data, the experimental findings show that the proposed methods surpassed current leading approaches based on several clustering validation metrics and reduced computational time significantly.
Dhaka Univ. J. Sci. 74(2): 271–282, 2026 (July)
The increasing volume, diversity, and complexity of Big Data require clustering techniques capable of handling high-dimensional, noisy, and heterogeneous datasets while maintaining scalability and robustness. This study proposes a BZ-Optimized K-Medoids Clustering Algorithm (BZ-KMedoids) that integrates fuzzy membership modelling, BZ-based medoid optimization, and distributed processing within a unified clustering framework. Fuzzy membership enables partial assignment of data points to multiple clusters, improving the representation of overlapping cluster boundaries and uncertain data regions. BZ optimization refines medoid selection using membership information, enhancing cluster compactness, stability, and resistance to outliers. Distributed execution further supports efficient large-scale processing through parallel computation across multiple nodes. The proposed framework addresses key limitations of conventional clustering approaches, including sensitivity to initialization, rigid cluster assignments, and reduced effectiveness in complex data environments. Experimental evaluation was conducted using the Bank Marketing dataset and compared with K-Means and Agglomerative clustering methods. The proposed approach achieved a clustering error of 0.38, substantially lower than the values of 3.55 and 3.59 obtained by K-Means and Agglomerative clustering, respectively. BZ-KMedoids also achieved an accuracy of 0.99, outperforming K-Means (0.92) and Agglomerative clustering (0.91). Although the average convergence time was 12.35 s compared with 4.87 s for K-Means, the improved clustering quality, robustness, and scalability provide a favorable trade-off between efficiency and performance. The proposed framework achieved superior clustering quality while maintaining scalability for large and heterogeneous Big Data environments. These findings demonstrate the effectiveness of BZ-KMedoids for adaptive, scalable, and high-quality clustering in practical analytics applications.
Chen-Xi Zheng, Qi-Chang Hu, Ruiqing Hong· International journal on art...· 0 citations
Clustering analysis is an essential task in data mining and machine learning, and the classical K-means algorithm is widely used due to its efficiency. However, its random initialization often leads to unstable results, especially on complex or nonuniform datasets, where it easily falls into local optima. Moreover, its objective function focuses solely on intra-cluster compactness while overlooking inter-cluster separability, thus limiting global clustering performance. To address these issues, this paper proposes an improved K-means method (MD-Kmeans). The algorithm integrates K-nearest-neighbor– based density estimation with a maximum-dispersion strategy to ensure representative and well-distributed initial centers, and employs a balanced objective that jointly enhances intra-cluster compactness and inter-cluster separability. Experimental results show that MD-Kmeans achieves notable improvements in Adjusted Rand Index (ARI), Silhouette Coefficient (SC), and Davies–Bouldin Index (DBI), outperforming traditional K-means and recent variants, particularly on non-uniform datasets.
Shaobo Deng, Yong-Kang Ye, Sujie Guan et al.· International Conference on...· 0 citations
In the Big Data era, the scalability of clustering algorithms constitutes a key challenge. Traditional density-based methods (e.g., DBSCAN) offer robustness to noise and the ability to detect non-linear clusters, yet their quadratic time complexity $O(N^2)$ drastically limits their applicability. Conversely, partitional algorithms (e.g., K-Means), with their linear complexity $O(N)$, impose sphericity on the resulting groups and fail in the presence of outliers. This paper presents K-SCAN -- a novel hybrid algorithm that optimizes this trade-off. The method integrates preliminary vector quantization (stochastic Mini-Batch K-Means) to extract a reduced set of weighted micro-clusters, followed by a subsequent density-based structural analysis. Empirical evaluation on datasets of up to $10^6$ samples confirms the linear computational complexity of the proposed solution. K-SCAN achieves more than a 3-fold speed-up over the hierarchical BIRCH algorithm, avoiding the costly management of tree-based structures. The method precisely identifies non-linear manifolds while maintaining structural stability (Adjusted Rand Index>0.99), even with noise levels reaching 55\% of the data volume. The main limitation of the proposed algorithm, which could not be fully eliminated in the present study, remains its susceptibility to over-smoothing and its difficulty in separating clusters with highly heterogeneous local density. In complex visual spaces, this can lead to the loss of the finest topological details.
F. Kosiorowski, Grzegorz Sroka· arXiv.org· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
