Data-Driven Discovery of Feature Groups in Clinical Time Series
This work addresses the problem of enhancing predictive modeling in healthcare by automating feature grouping, which is incremental as it builds on existing deep learning architectures with a novel clustering approach.
The paper tackles the challenge of defining feature groups in clinical time series data by proposing a method that learns these groups through clustering of embedding layer weights, which improves downstream performance on clinically relevant tasks, achieving results comparable to expert-defined groups on real-world data.
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.