Machine Learning Algorithm-Based Prediction of Hyperglycemia Risk After Acute Ischemic Stroke

The importance of early prediction of hyperglycemia after acute ischemic stroke in intensive care unit (ICU) was addressed to explore the application of machine learning methods in disease prediction. This research focuses on the prediction of the risk of causing hyperglycemia using data from acute ischemic stroke patients in the MIMIC (Medical Information Mart for Intensive Care) - IV database. We construct and compare 8 machine learning models of Logistic Regression, Support Vector Machines, Decision Trees, Random Forests, Gradient Boosting, Bayesian Classifiers, K-Nearest Neighbors, and XGBoost. Feature selection and parameter tuning were performed to improve the model performance and to find the optimal prediction model for hyperglycemia after acute ischemic stroke. After comparison, the support vector machine-based model had the best prediction accuracy compared to other models, with accuracy, precision recall, and f1 score reaching 97.84%, 97%, 98%, and 97%, respectively. The machine learning method has high accuracy in the early identification of hyperglycemia after acute ischemic stroke in ICU, which is expected to be an objective and effective decision-making tool to assist clinicians.