Subgraph Propagation and Contrastive Calibration for Incomplete Multiview Data Clustering

Clustering uncertain overlapping symptoms of multiple diseases in clinical diagnosis

In various fields, including medical science, datasets characterized by uncertainty are generated. Conventional clustering algorithms, designed for deterministic data, often prove inadequate when applied to uncertain data, posing significant challenges. Recent advancements have introduced clustering algorithms based on a possible world model, specifically designed to handle uncertainty, showing promising outcomes. However, these algorithms face two primary issues. First, they treat all possible worlds equally, neglecting the relative importance of each world. Second, they employ time-consuming and inefficient post-processing techniques for world selection. This research aims to create clusters of observed symptoms in patients, enabling the exploration of intricate relationships between symptoms. However, the symptoms dataset presents unique challenges, as it entails uncertainty and exhibits overlapping symptoms across multiple diseases, rendering the formation of mutually exclusive clusters impractical. Conventional similarity measures, assuming mutually exclusive clusters, fail to address these challenges effectively. Furthermore, the categorical nature of the symptoms dataset further complicates the analysis, as most similarity measures are optimized for numerical datasets. To overcome these scientific obstacles, this research proposes an innovative clustering algorithm that considers the precise weight of each symptom in every disease, facilitating the generation of overlapping clusters that accurately depict the associations between symptoms in the context of various diseases.