The electrocardiogram (ECG) is a crucial non-invasive diagnostic tool for cardiovascular disease (CVD), the leading cause of death globally. Clinical expertise is necessary for ECG interpretation; However, it takes time and is subjective. The objective of this study is to improve the accuracy and effectiveness of CVD detection by computerizing ECG diagnosis using machine learning (ML) and deep learning (DL) techniques, more especially convolutional neural networks, or CNNs. Our approach makes use of an emerging CNN framework, along with pretrained models such as SqueezeNet and AlexNet, in the classification process of four predominant heart conditions: normal person (NP), abnormal heartbeat (AH), myocardial infarction (MI), and history of myocardial infarction (H.MI). ECG images are preliminarily processed and augmented by data to make them uniform for improving dataset balancing. The images are then sent through our presented CNN for the purpose of extracting features and their classification. CNN and pretrained model features are also used in conjunction with more conventional machine learning classifiers, such as Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree (DT), Random Forest (RF), and Naive Bayes (NB). The proposed CNN model attained 98.23% accuracy, but using its features with NB resulted in an amazing 99.79% accuracy. This is a combination of the strength of deep feature extraction and the interpretability of traditional ML classifiers. The model holds potential for real-time, computerized cardiovascular screening and can be of immense assistance in clinical decision-making. Hyperparameter optimization and generalizing the model to other biomedical imaging tasks could be interesting areas of future work.