Optimized hybrid RNN-GRU model for predictive diagnosis of cardiovascular disease.

Abstract

Cardiovascular disease (CVD) continues to be the leading cause of death for individuals all over the globe, and India bears a disproportionate share of the burden associated with this condition. A hybrid deep learning model that combines Recurrent Neural Networks (RNN) and Gated Recurrent Units (GRU) is being used in this research project with the objective of enhancing the accuracy and efficiency of heart disease risk prediction. It makes use of a dataset consisting of 918 samples that was obtained from IEEE Dataport. It then applies preprocessing processes such as the correction of outliers using the Interquartile Range (IQR) technique and the normalization of numerical characteristics. The use of Synthetic Minority Over Sampling Technique (SMOTE) to get a balanced dataset, the dataset is then divided into training and testing sets. For the purpose of fine-tuning the model, GridSearchCV was used in conjunction with 10-fold cross-validation. The results demonstrated that the hybrid RNN-GRU model greatly outperformed the performance of the separate RNN and GRU models. It achieved an accuracy of 99.6%, a 99.6% F1 score, a 99.6% precision, and a 99% recall, which was higher than the highest reported model accuracies of 87% and 97%. The results of this study demonstrated that the capacity of RNNs to process sequences, when paired with the gating properties of GRUs, allows the extraction of temporal parameters from cardiac signals. The significance of appropriate data processing highlights the potential contribution of the model to clinical decision-making procedures that are targeted at early and more accurate detection of cardiac disease.