Detection of Bloodstream Infections: A Comparative Analysis of Machine Learning Algorithms

Bloodstream infections (BSIs) represent a critical threat to the health of hemodialysis patients, necessitating timely and precise detection for effective intervention and improved clinical outcomes. This paper explores the utilization of machine learning models for predicting BSIs in hemodialysis patients. Multiple algorithms, including Naive Bayes, Support Vector Machines (SVM), K-Nearest Neighbors (KNN), XGBoost, LightGBM, and Random Forest, were employed to analyze patient datasets. The models were assessed based on their classification accuracy. To further enhance performance, ensemble techniques―such as bagging, boosting, stacking, and voting―were applied, along with hyperparameter optimization using methods like grid search, randomized search, and Bayesian optimization. The study found that ensemble models and optimized hyperparameters yielded significant improvements in prediction accuracy, with XGBoost and Random Forest consistently achieving superior performance, reaching an accuracy of up to 79.8%. These findings underscore the potential of machine learning algorithms to aid in the early detection of BSIs, offering a valuable tool for improving patient care and reducing the associated health risks.