aspg-18-2908-1792025536 20261015005216 American Scientific Publishing Group admin@americaspg.com American Scientific Publishing Group Journal of Intelligent Systems and Internet of Things JISIoT 2769-786X 2690-6791 2024 13 1 Parvathy S. Department of CSE, Vels Institute of Science, Technology and Advanced Studies, Chennai, India A. Packialatha Department of CSE, Vels Institute of Science, Technology and Advanced Studies, Chennai, 600117, India Recently, Heart diseases is considered as the one of deadliest diseases which has resulted in the increased death rates across the globe. Predicting heart diseases requires vast experiences along with advanced knowledge. IoT and AI are two emerging technologies that help in heart disease prediction. High diagnostic accuracy with minimal processing overhead, however, continues to be a design problem for researchers. To address this problem, this paper develops the Intelligent IoT structure for the better prediction of cardiac diseases employing Harris Hawk Optimized Gated Modified Recurrent Units (HHO-M-GRU). The paper also proposes the real time data collection using IoT wearable test beds which comprises of electrocardiography sensors (ECG) interfaced with MICOTT Boards & ESP8266 transceivers. For later processing, the acquired data are saved on the cloud. The proposed deep learning network is utilized for evaluating the received heart data and used for predicting the heart diseases. Additionally, the suggested HHO-GRU is trained with the versatile datasets which consist of normal and abnormal stages of heart diseases. By calculating the suggested model's performance measures, including accuracy, precision, recall, specificity, and F1-score, a thorough experiment is conducted. The proposed framework was implemented in Keras libraries with Tensorflow 2.1.1 as backend. Furthermore, prediction performance and complexity overhead is compared using the other cutting-edge deep learning algorithms already in use to demonstrate the model's superiority. in predicting the heart diseases. The suggested approach beats previous models for learning with respect to of accurate prediction (99%) and minimal computing overhead, according to the results. 2024 259 275 2908 https://creativecommons.org/licenses/by/4.0/ 10.54216/JISIoT.130119 https://americaspg.com/journal/18/article/2908 https://americaspg.com/storage/21719248102.pdf