Clustering and Interpretation of Time-series Trajectories of chronic pain using Evidential c-means (Conference)

Abstract

The most well-known unsupervised classification algorithms allow for the identification of hard or probabilistic partitions. However, in complex healthcare datasets, these algorithms may have limitations in capturing uncertainty and handling outliers or imprecise observations. The aim is to analyze time series data of patients with chronic pain and identify distinct care trajectories. A fuzzy clustering approach is proposed based on feature extraction and feature selection, aiming to improve interpretability and enhance the performance of the clustering procedure. The initial step involves extracting features from time series data and selecting essential attributes using Tsfresh and unsupervised feature selection methods like unsupervised random forest, Laplacian score, and unsupervised spectral feature selection. The second step involves using the evidential c-means (ECM) clustering algorithm on the extracted attributes. ECM method based on belief functions, allows for generating a credal partition that can model various forms of uncertainty. The results reveal the existence of two clusters of chronic pain related to discomfort and well-being, exhibiting excellent separability and compactness. Additionally, an uncertain cluster groups patients with intermediate characteristics. Interpreting the partitions involved descriptive analysis, statistical tests, and multinomial regression on clinical and demographic data to understand patient profiles within identified trajectories

Download paper here

Armel Soubeiga , Violaine Antoine, Alice Corteval, Nicolas Kerckhove, Sylvain Moreno, and Issam Falih, " Clustering and Interpretation of Time-series Trajectories of chronic pain using Evidential c-means," 2023, 6th International Conference of the ERCIM WG on Computational and Methodological Statistics, CFE-CMStatistics 2023