Survival Prediction of Septic Patients Using Boosting and Oversampling Techniques

Sepsis can pose a serious threat to the human body. Being a rapidly progressing condition, it could lead to death even within a few hours of developing, contributing to a huge chunk of global deaths, which is incredibly alarming. Hence predicting the likelihood of death among affected patients would give us an upper hand in treating the death prone ones. And thus we have chosen a dataset containing information about age, gender, number of septic episodes and finally the survival status of patients gathered from hospitals. Due to a high range of imbalance within the dataset, we have performed oversampling using SMOTE and trained six models using the Machine Learning Classifiers: Logistic Regression, Support Vector Machine, Random Forest, Decision Tree, K-Nearest Neighbors and Naive Bayes. As the results received were not satisfactory, we opted out for boosting techniques and used XGBoost, AdaBoost and Gradient Boosting which gave us better results.