Table 1.
The 10-features considered during this study.
Fig 1.
Architecture of the proposed approach.
Fig 2.
The process of Over-sampling and Under-sampling techniques.
Fig 3.
The step by step process of stratified approach with cross validation.
Table 2.
Hyper-parameters used in different classifiers.
Fig 4.
An example of confusion matrix.
0 denotes patients with normal DE-MRI, 1 patients with Myocardial Infarction, and 2 patients with Myocarditis.
Table 3.
Accuracy comparison obtained with all combinations.
Best result in bold. (Stratified denoted as STRF).
Table 4.
Execution time of all combinations for training (304 cases) and validation (25 cases)—in seconds.
(Stratified denoted as STRF).
Fig 5.
Wrapper method for feature importance identification.
Table 5.
Feature importance for the classification (using Random Forest).
Fig 6.
Comparison of accuracy of Support Vector Machine (SVM) classifier, K-Nearest Neighbors (KNN), Random Forest (RF), Extremely Randomised Tree (ERT), Gradient Boosting (GB), Decision Tree (DT), Multi-Layer Perceptron (MLP), eXtreme Gradient Boost (XGB), Light Gradient Boost Machine (LGBM) and Stacked generalization (Stacking).
(a) Accuracy distribution of Stratified method (10-fold cross-validation), (b) Accuracy distribution of Stratified and Under-sampling (10-fold cross-validation), (c) Accuracy distribution of Stratified and Over-sampling (10-fold cross-validation), (d) Accuracy distribution of Stratified and NearMiss (10-fold cross-validation, (e) Accuracy distribution of Stratified and SMOTE (10-fold cross-validation).
Fig 7.
Comparison of the confusion matrices for the different approaches.
0 denotes patients with normal DE-MRI, 1 patients with Myocardial Infarction, and 2 patients with Myocarditis. (a) Confusion matrix for LGBM (OS). (b) Confusion matrix for LGBM (SMOTE). (c) Confusion matrix for Stacking (OS). (d) Confusion matrix for Stacking (SMOTE).
Table 6.
Metrics calculated from the confusion matrices.