Liver Disease Detection using Hybrid Unsupervised Techniques: A Model-Level Comparative Study

Liver-related health issues are often diagnosed late due to the absence of early symptoms and the complexity of medical datasets. This research investigates the potential of hybrid unsupervised learning methods to support early detection by uncovering patterns in unlabeled liver disease data. We combine dimensionality reduction techniques, such as PCA and t-SNE, with clustering models like K-Means, DBSCAN, and hierarchical clustering to improve data grouping and insight extraction. A model-level comparison is carried out using internal validation metrics and expert-labeled data references. The outcomes indicate that integrated approaches outperform standalone algorithms in generating meaningful clusters, with the PCA-hierarchical combination demonstrating the highest level of coherence. These findings underscore the value of hybrid unsupervised models as practical tools for early, non-invasive liver disease assessment using raw clinical data.