Find in Library

Myocardial Infarction, commonly known as heart attack, is an extremely dangerous condition caused by inadequate blood flow to the heart. It is a major cause of death worldwide and swift appropriate actions need to be taken after hospitalization to save a patient's life. Clinicians need to be extremely cautious in handling such situations. A decision-support system predicting the risk in MI patients can be quite beneficial to clinicians. In this study, we develop such a system to predict the complications in MI patients immediately after hospitalization. In that regard, we utilized a dataset containing the records of 1700 MI patients. Class imbalance is a major concern in most medical datasets as it biases the predictions of the ML algorithms towards the majority (negative) class. Appropriate measures need to be taken to address the imbalance scenario and develop a reliable prediction system. In that regard, we propose a new approach that combines sampling techniques with cost-sensitive learning. The advantage of this hybrid approach is that it reduces the number of minority class samples required to be generated or the number of majority class samples needed to be eliminated to obtain balance. Additionally, it reduces the weight to be assigned as the penalty for the cost-sensitive classifier. The complications associated with using sampling or cost-sensitive learning separately are reduced by using this kind of hybridization. We hypothesize that a suitable balance between the two approaches can optimize the prediction performance. The proposed approach performed significantly better than both traditional sampling techniques and cost-sensitive learning. The highest ROC-AUC score of 80.88% and MCC score of 66.53% were achieved using this approach. External validation was also performed on 36 imbalanced datasets and the proposed methodology outperformed other popular techniques used in imbalanced learning. Thus, the prediction framework presented in this study can ensure reliable risk prediction of MI patients at an early stage of hospitalization which can be quite beneficial to clinicians in the decision-making process.