Myocardial infarction (MI) remains one of the leading causes of cardiac-related mortality globally, posing a significant burden on public health systems, including in India. Early and accurate diagnosis is crucial for the effective management and treatment of MI and its various subtypes. Electrocardiogram (ECG) serves as a primary, non-invasive diagnostic tool due to its capacity to reveal critical cardiac anomalies, particularly through analysis of the QRS complex. However, reliable interpretation of ECG signals is challenged by artifacts such as baseline wander, motion interference, and muscle contraction noise. The study explores about spectral techniques such as Stationary Wavelet Transform (SWT) to decompose ECG signals into relevant timefrequency domains. From these decomposed spectral bands, nonlinear entropy-based features are extracted to capture the inherent complexity and irregularities within the MI signal dynamics. These features are then employed to train multiple machine learning classifiers including Random Forest (RF), Bayesian Optimized Support Vector Machine (BOSVM), Adaptive K-Nearest Neighbours (AKNN), SVM (Support Vector Machine), AdaBoost, Decision Tree (DT), and Extreme Learning Machine (ELM). The simulation results indicated that the combined SWT and Sample Entropy and Bubble Entropy approach achieved an accuracy of 99.51%, specificity of 99.63%, sensitivity of 98.72%, precision of 98.6%, and F1-score of 98.59%.