Data-Driven Discovery of Feature Groups in Clinical Time Series

2025-11-11

AI Summary

Clinical time-series data exhibit abundant heterogeneous features with limited prior semantic knowledge, hindering task-oriented feature grouping. To address this, we propose a data-driven, end-to-end feature clustering method: a supervised learning framework jointly optimizes feature embedding weights and performs differentiable clustering on the embeddings to automatically discover dynamic feature groups highly relevant to the prediction task. The approach requires no domain-specific priors and simultaneously improves model performance and clinical interpretability. On synthetic benchmarks, it significantly outperforms static clustering baselines; on real-world intensive care unit data, it achieves predictive accuracy comparable to expert-defined groupings and uncovers clinically meaningful variable associations—e.g., coordinated dynamics between blood pressure and heart rate. Our key innovation lies in integrating embedding-weight clustering as a fully differentiable module within the supervised training pipeline, enabling task-aware discovery of latent feature structure.

Technology Category

• Machine Learning: Clustering
• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Normalization, clustering, classification, and summarization of Web text

Abstract

Clinical time series data are critical for patient monitoring and predictive modeling. These time series are typically multivariate and often comprise hundreds of heterogeneous features from different data sources. The grouping of features based on similarity and relevance to the prediction task has been shown to enhance the performance of deep learning architectures. However, defining these groups a priori using only semantic knowledge is challenging, even for domain experts. To address this, we propose a novel method that learns feature groups by clustering weights of feature-wise embedding layers. This approach seamlessly integrates into standard supervised training and discovers the groups that directly improve downstream performance on clinically relevant tasks. We demonstrate that our method outperforms static clustering approaches on synthetic data and achieves performance comparable to expert-defined groups on real-world medical data. Moreover, the learned feature groups are clinically interpretable, enabling data-driven discovery of task-relevant relationships between variables.

Problem

• Automatically grouping clinical time series features to improve prediction performance
• Overcoming challenges in manual feature grouping using semantic knowledge alone
• Learning clinically interpretable feature groups through weight clustering in embedding layers

Innovation

• Learns feature groups by clustering embedding weights
• Integrates into standard supervised training process
• Produces clinically interpretable data-driven feature relationships