An Ensemble Machine Learning Model for the Early Detection of Sepsis from Clinical Data

Mengsha Fu1, Menglin LU2, Zhenhua Xu3
1Nanjing University Of Aeronautics And Astronautics, 2Dalian University of Technology, 3Beijing Healsci Technology Co


Abstract

An Ensemble Machine Learning model for the early detection of sepsis from Clinical Data Mengsha F Menglin L Zhenhua X Aims: The goal is the early detection of sepsis using physiological data. The study aimed to predict sepsis 6 hours before the clinical prediction of sepsis, and there is a utility function that rewards early predictions and penalizes late predictions as well as false alarms. Method: First analyzed and preprocessed the data, included: excluding outliers and filling missing values with the mean value. Patients were excluded if they developed sepsis with the first record and age < 18 years; Features were chosen from 40 columns, Variables with too large missing values and low correlation are excluded. An Integrated model was designed to train and predict. Result´╝Ü5000 files from training_setB were used to train, where included 279 patients,188453 records, 2623 sepsis records; The test data were randomly chosen from training_setA:2000 files, 184 patients,1760 sepsis records; The utility score is 0.81 in the local test files.Additional evaluation indicators (AUROC=0.74,AUPRC=0.77). Conclusion: An Ensemble model can predict sepsis 6 hours prior to its onset. Future studies will validate in a large cohort.