Importance of early prediction of Sudden Cardiac Deaths (SCD) has been rising as a large percentage of mortality of patients with cardiovascular diseases. Various deep learning methodologies has been developed to predict the onset of SCDs, Their key limitation is either classification accuracy or the processing time. This research tries to improve the classification accuracy and decrease the processing time. A Convolutional Neural Network (CNN) is combined with a Recurrence Complex Network (RCN) along with Dropout Regularization to enhance the accuracy of SCD classification. Initially, the synchronization feature of individual heartbeat of the electrocardiogram (ECG) signal is constructed by RCN. The recurrence matrix from the (RCN) will generate Eigen values. Then, CNN will be employed to extract features and detect SCD by analysing the Eigen values. Finally, the performance of the classification is improved by the developing a voting algorithm for the SCD detection. MIT-BIH SCD database is used to evaluate the proposed system. The average accuracy and processing time for MIT-BIH Arrhythmia dataset is 93.24% and 21 epochs, MIT-BIH SCD Holter dataset is 90.60% and 11.5 epochs, and Apnoea-ECG dataset is 92.13% and 13.5 epochs. The average processing time has also been reduced to 20.77 milliseconds against the current processing time of 32.96 milliseconds. The proposed system enhances the classification accuracy and the processing time of the prediction system. The study eradicates the issue of gradient saturation during the training of the CNN by proposing a new activation function as well as eliminates the risk of overfitting by implementing dropout regularization in CNN.