The Effectiveness of Data Imputations on Myocardial Infarction Complication Classification Using Machine Learning Approach with Hyperparameter Tuning

Abstract

Complications from Myocardial Infarction (MI) represent a critical medical emergency caused by the blockage of blood flow to the heart muscle, primarily due to a blood clot in a coronary artery narrowed by atherosclerotic plaque. Diagnosing MI involves physical examination, electrocardiogram (ECG) evaluation, blood sample analysis for specific heart enzyme levels, and imaging techniques such as coronary angiography. Proactively predicting acute myocardial complications can mitigate adverse outcomes, and this study focuses on early prediction using classification methods. Machine learning algorithms such as Support Vector Machine (SVM), Random Forest, and XGBoost were employed to classify patient medical records accurately. Techniques like K-Nearest Neighbors (KNN) imputation, Iterative imputation, and Miss Forest were used to handle incomplete datasets, preserving vital information. Hyperparameter optimization, crucial for model performance, was performed using Bayesian Optimization, which minimizes the objective function by modeling past evaluations. The contribution to this study is to see how much influence data imputation has on classification using machine learning methods on missing data and to see how much influence the optimization method has when performing hyperparameter tuning. Results demonstrated that the Iterative Imputation method yielded excellent performance with SVM and XGBoost algorithms. SVM achieved 100% accuracy, precision, sensitivity, F1 score, and AUC. XGBoost reached 99.4% accuracy, 100% precision, 79.6% sensitivity, an F1 score of 88.7%, and an AUC of 0.898. KNN Imputation with SVM showed results similar to Iterative Imputation with SVM, while Random Forest exhibited poor classification outcomes due to data imbalance, causing overfitting.