Performance Analysis of Class Imbalance Handling Techniques for Early Sepsis Prediction using Machine Learning Algorithms

Abstract

Sepsis is a life-threatening condition that may cause mortality in ICU patients. Researchers have formerly developed Machine Learning (ML) models to predict Sepsis and other medical conditions. Performance of ML models dependent on the quality of the data set used for training. Class imbalance is one of the problems observed in medical data sets. In this paper, we have presented a comparative study of three class imbalance handling techniques to identify the best among the three. We made a comparison on performance metrics of predictive models developed using ML classification algorithms. We proposed a four-phase approach for predictive model development. Consequently, we used healthcare parameters available at any peripheral basic health facility and do not require laboratory investigations. Further, we proposed a novel machine learning predictive model, “Sepsis Prediction Model for Peripheral Hospitals (SPMPH), ” for sepsis prediction in ICU patients. SPMPH provides better accuracy (.95), precision (.98), recall (.91) and AUROC (.978) as compared to other available models.