Using a realistic dataset based on the United Kingdom clinical exercise studies Datalink, various architectures are evolved and evaluated to decide the simplest for predicting the risk of cardiac arrest. The architectures examined include several modified lengthy short-term memory networks, multi-input/multi-output architectures, Convolutional Neural Networks (CNNs), and Residual Neural Networks. Moreover, the consequences of the architectures on the predictions are compared while the statistics are supplemented with external capabilities. Furthermore, the results recommend increasing the facts with external functions, enhancing prediction accuracy for several architectures. Via this observation, conclusions may be drawn about which architectures are satisfactory-desirable for predicting cardiac arrest chance, as well as the consequences of external capabilities on the effectiveness of the architectures.