Machine Learning Models for Early Prediction of Malignancy in Sepsis Using Clinical Dataset

Research on predicting sepsis using machine learning algorithms is dramatically increasing due to their good performance. Moreover, sepsis or its malignancy is life-threatening, that is contributing to 5.3 million deaths worldwide. A meta-analysis estimated about 31.5 million sepsis and 19.4 million severe sepsis cases occur each year. Screening bulk numbers correctly is tedious, time-consuming, and subjective in nature. Automatic methods for early prediction of sepsis and malignancy will assist doctors in treatment planning or surgical planning. This study investigated five different machine learning models for predicting sepsis as benign or malignant. The most commonly and publicly available PhysioNeT 2019 sepsis challenge dataset is used to train these five models. The accuracy of a logistic regression model is 0.76, the Naive Bayes classifier is 0.75, KNN classifier is 0.83, the XGboost classifier is 0.86, and the Random Forest classifier is 0.96 for 9171 test samples. The Random Forest classifier has performed the best among the five due to its ensemble learning approach.