Using Supervised Feature Selection Methods to Improve the Predictive Performance of Clinical Outcomes in Intensive Care Units
Document Type
Presentation
Publication Date
2023
Publication Title
IISE Annual Conference & Expo 2023
Pages
2774
Conference Name
IISE Annual Conference & Expo 2023, May 20-23, 2023, New Orleans, Louisiana
Abstract
Better prediction of clinical outcomes can help reduce costs and improve the success rate of surgery. Identifying the essential features from a medical dataset to conduct machine learning analysis could be crucial to achieving better prediction performance. This research compares the performance of several supervised feature selection methods in a highly imbalanced dataset to predict mortality in patients undergoing ICU heart surgery. Feature selection is the process of reducing the number of input variables when developing a predictive model, which is an ensemble stacking ML model in this research. This approach is desirable to reduce the computational cost of modeling and, in some cases, improve the model's performance by providing an appropriate subset of features as input for ML models. Data sets used to predict mortality in heart surgery patients are very unbalanced, which can lead to model overfitting. To address this problem, we propose a combination of resampling methods to improve the prediction performance of the minority and majority classes using a large dataset. Data were collected from the hospitals related to Shahid Beheshti University of Medical Sciences and Health Services in Iran between 2015 and 2020 during and after a heart surgery (recovery) process for hospitalized patients. Initial results show that each Wrapper, Filter, and Embedded feature selection methods select a different essential subset of features. This research compares their effect on a stacking ensemble ML model to select the most efficient subset of features to achieve better mortality prediction performance.
Rights
Copyright 2023 Institute of Industrial and Systems Engineers.
All rights reserved. Included with the kind written permission of the copyright holders.
ORCID
0009-0003-9798-8958 (Ghavidel), 0000-0003-4348-7798 (Pazos)
Original Publication Citation
Ghavidel, A., & Pazos-Lago, P. (2023). Using supervised feature selection methods to improve the predictive performance of clinical outcomes in Intensive Care Units [Oral Presentation]. IISE Annual Conference & Expo 2023, New Orleans, Louisiana. https://iise.confex.com/iise/2023/meetingapp.cgi/Paper/2774
Repository Citation
Ghavidel, Arman and Pazos-Lago, Pilar, "Using Supervised Feature Selection Methods to Improve the Predictive Performance of Clinical Outcomes in Intensive Care Units" (2023). Engineering Management & Systems Engineering Faculty Publications. 255.
https://digitalcommons.odu.edu/emse_fac_pubs/255