Cardiovascular disease (CVD) continues to be one of the leading causes of death worldwide, with approximately 20 million deaths taking place in 2015. Studies show that deaths from CVD will amount to 23.6 million annually by 2030. Early recognition is vital to improve patient survival rates, particularly in the health system where there are limited resources and insufficiency of experienced professionals. The objective of this study was to develop an automated machine learning technique which would classify paper electrograph (ECG) recordings into four electrocardiographically different classes, namely acute myocardial infarction (MI), previous MI, irregular cardiac rhythm, and normal cardiac function. Using the broad ECG Images DataSet on Kaggle with cardiac patient data (i.e., 11,148 images processed and augmented) three different convolutional neural network (CNN) models were tested: VGG16, MobileNetV2 and InceptionV3. Of those VGG16 gave the best result overall by balanced performance measures, achieving 98.71% classification accuracy and F1 score 0.9853 in respect of results. This was a marked improvement when comparing results obtained with the InceptionV3 and MobileNetV2 models especially for classes of cases less well represented within these data, such as previous myocardial infarction cases. This is encouraging for the use of systems based on CNNs allowing rapid and effective interpretation of ECGs thereby guiding rapid diagnosis and reducing the risk of fatal outcome.