ABSTRACT
One of the crucial domains in healthcare research is the timely prediction of cardiac arrest from electrocardiogram signals. Developing a robust predictive framework should help improve patient outcomes and reduce the mortality rates of associated cardiovascular diseases. In this study, we apply state-of-the-art machine learning methods, such as artificial neural networks, SVMs, RF, and LR to develop such a framework. Our solution is based on distributed deep learning and takes advantage of cloud computing to process large-scale electrocardiogram datasets obtained for 4320 patients in conjunction with their health records. The cloud computing approach should enhance the accuracy and scale of predictive models, ensuring the timely detection of patient states that might lead to cardiac events. In addition to these advantages offered by the research framework, we also contribute to the analysis with novel techniques of ensemble learning and deep learning methods based on CNNs and RNNs. The purpose of these approaches is to better capture untargeted patterns in the electrocardiogram data and increase the precision of predictive models. The comparative performance test demonstrated the superiority of our model only after vigorous training and testing in terms of recall, F1 score, and accuracy in addition to impressive precision values. Another important aspect that we take into consideration is that IoT allows monitoring patients’ states in real-time and, consequently, building predictive models to be more personalized. Our research provides the development of these predictive healthcare analytics and should lead to advances in both cardiac care and healthcare.
