| Summary: | Multi-view clustering (MVC) has gained considerable attention for its ability to integrate diverse representations of data, thereby enhancing clustering performance over traditional single-view techniques. However, constraints are still encountered including view inconsistency, high dimensionality and difficulty in integrating complementary information from heterogeneous data sources. This research addresses these limitations through the development of two enhanced methods, Multi-View Co-Clustering with Feature Selection (MVCCFS) and Exponential Decay-based Multi-View Co-Clustering (ED-MVCC). The MVCCFS method introduces a rank-based feature selection mechanism that identifies and retains the most informative features from each view and computational burden while improving cluster interpretability and accuracy. The ED-MVCC method incorporates an exponential decay-based dynamic weighting strategy within a bipartite graph framework, emphasizing coherent inter-view relationships and suppressing noisy or conflicting information. Additionally, a Dual-Threshold Consensus strategy is proposed to resolve agreement and disagreement among views, enhancing the robustness of the co-clustering process. Extensive experiments on six benchmark datasets demonstrate that the proposed methods outperform eight state-of-the-art multi-view clustering techniques, achieving up to 18% improvement in clustering accuracy. The approaches also show superior performance in terms of precision. This study significantly contributes to the field by offering scalable, adaptive, and interpretable co-clustering solutions for complex, high-dimensional, and heterogeneous multi-view datasets.
|
|---|
