Objective: Sudden cardiac death (SCD) remains one of the leading causes of mortality worldwide. The purpose of this study is to predict SCD early and to distinguish it from congestive heart failure (CHF), which presents with similar electrocardiographic patterns. Research Method: In this study, shallow and fully automated models based on convolutional neural networks (CNNs) are presented to predict SCD, so that the electrocardiogram (ECG) signal is given as input to the model, and classification is performed at the end of the model. Results: The first model, which is based on one-dimensional CNNs, was able to predict SCD with an accuracy of 96%. In the next step, by adding data related to CHF, an accuracy of 97.85 was obtained in the three-class classification (normal, before sudden cardiac death, and heart failure). In the second model, which is based on two-dimensional CNNs, first, each part of the ECG signals was converted into two-dimensional images using continuous wavelet transform (CWT) and given as input to the model. This model was able to predict SCD with 99% accuracy. Conclusion: The proposed fully automated CNN-based models demonstrated high predictive accuracy for SCD within one hour before its occurrence.