Article: Screening and sparsifying maternal immune features for predicting labour induction based on a glass-box model Journal: International Journal of Data Mining and Bioinformatics (IJDMB) 2025 Vol.29 No.6 pp.1 - 15 Abstract: Immune signatures strongly associate with the progression of labour towards the active phase of labour. However, the detailed relationship is still not clear. Herein, interpretable machine learning methods are implemented for mining complex immune data. Principal component analysis and covariance analysis are employed to achieve dimensionality reduction of the immune features (1,058) as input. Using 16 key immune features as input, RMSE decreased from 277 min to 214 min by Ridge model. Moreover, sure independence screening and sparsifying operator (SISSO) was implemented to establish a glass-box model for generating interpretable mathematical information format of key immune features associated with induced labour progression. The prediction accuracy was further improved by SISSO input with only 14 features (R<SUP align="right"><SMALL>2</SMALL></SUP> = 0.9934, RMSE = 42 min, MAE = 30 min), and the exact mathematical format of the model was obtained [equation (5)]. Reliable description of progression is established from labour induction until establishing active labour. Inderscience Publishers - linking academia, business and industry through research

Title: Screening and sparsifying maternal immune features for predicting labour induction based on a glass-box model

Authors: Rong Hu; Jing Li; Yan Liu; Qianqian Zhang; Zhaoyi Bai; Zhongwei Zhao; Weiwei Guo; Rong Liu

Addresses: Department of Obstetrics and Gynaecology, Tianjin First Central Hospital, #24 Fukang Road, Nankai District, Tianjin 300192, China ' Department of Obstetrics and Gynaecology, Tianjin First Central Hospital, #24 Fukang Road, Nankai District, Tianjin 300192, China ' Department of Obstetrics and Gynaecology, Tianjin First Central Hospital, #24 Fukang Road, Nankai District, Tianjin 300192, China ' Department of Obstetrics and Gynaecology, Tianjin First Central Hospital, #24 Fukang Road, Nankai District, Tianjin 300192, China ' Department of Obstetrics and Gynaecology, Tianjin First Central Hospital, #24 Fukang Road, Nankai District, Tianjin 300192, China ' Department of Obstetrics and Gynaecology, Tianjin First Central Hospital, #24 Fukang Road, Nankai District, Tianjin 300192, China ' Department of Obstetrics and Gynaecology, Tianjin First Central Hospital, #24 Fukang Road, Nankai District, Tianjin 300192, China ' Department of Obstetrics and Gynaecology, Tianjin First Central Hospital, #24 Fukang Road, Nankai District, Tianjin 300192, China

Abstract: Immune signatures strongly associate with the progression of labour towards the active phase of labour. However, the detailed relationship is still not clear. Herein, interpretable machine learning methods are implemented for mining complex immune data. Principal component analysis and covariance analysis are employed to achieve dimensionality reduction of the immune features (1,058) as input. Using 16 key immune features as input, RMSE decreased from 277 min to 214 min by Ridge model. Moreover, sure independence screening and sparsifying operator (SISSO) was implemented to establish a glass-box model for generating interpretable mathematical information format of key immune features associated with induced labour progression. The prediction accuracy was further improved by SISSO input with only 14 features (R² = 0.9934, RMSE = 42 min, MAE = 30 min), and the exact mathematical format of the model was obtained [equation (5)]. Reliable description of progression is established from labour induction until establishing active labour.

Keywords: interpretable machine learning; labour induction; compressed-sensing method; regression; pregnancy.

DOI: 10.1504/IJDMB.2025.148381

International Journal of Data Mining and Bioinformatics, 2025 Vol.29 No.6, pp.1 - 15

Received: 22 Feb 2025
Accepted: 28 Jul 2025
Published online: 02 Sep 2025 *

Title: Screening and sparsifying maternal immune features for predicting labour induction based on a glass-box model

Keep up-to-date