Institution profile

Université Lumière Lyon 2

Academic institutioneurope · fr
Official website
Research library1linked papers
Opportunities0open roles
Selected work

Representative Papers

User Profiles of Sleep Disorder Sufferers: Towards Explainable Clustering and Differential Variable Analysis

Oct 13, 2025

Sleep disorders exhibit high clinical heterogeneity, impeding accurate diagnosis. To address this, we propose a patient stratification framework integrating explainable artificial intelligence (XAI) with unsupervised clustering. First, an enhanced clustering algorithm is applied to real-world anonymized clinical data to identify latent sleep disorder subtypes. Second, XAI techniques—including SHAP and LIME—are employed to interpret subtype-specific drivers, while differential feature analysis quantifies the contribution of each clinical variable. This approach unifies clustering transparency with model interpretability, substantially improving clinical readability and verifiability of algorithmic decisions. Experimental results demonstrate robust identification of clinically distinct subtypes characterized by divergent phenotypic profiles and risk factor patterns. The framework thus provides interpretable computational support for precision phenotyping, pathophysiological investigation, and personalized intervention design.

0 citationsRead paper
Recent publications

Latest Papers

User Profiles of Sleep Disorder Sufferers: Towards Explainable Clustering and Differential Variable Analysis

Oct 13, 2025

Sleep disorders exhibit high clinical heterogeneity, impeding accurate diagnosis. To address this, we propose a patient stratification framework integrating explainable artificial intelligence (XAI) with unsupervised clustering. First, an enhanced clustering algorithm is applied to real-world anonymized clinical data to identify latent sleep disorder subtypes. Second, XAI techniques—including SHAP and LIME—are employed to interpret subtype-specific drivers, while differential feature analysis quantifies the contribution of each clinical variable. This approach unifies clustering transparency with model interpretability, substantially improving clinical readability and verifiability of algorithmic decisions. Experimental results demonstrate robust identification of clinically distinct subtypes characterized by divergent phenotypic profiles and risk factor patterns. The framework thus provides interpretable computational support for precision phenotyping, pathophysiological investigation, and personalized intervention design.

0 citationsRead paper