Predictive Diagnosis of Cardiovascular Disease Using an Optimized RNN- GRU Hybrid Model

Cardiovascular disease (CVD) continues to be the foremost cause of mortality worldwide, claiming millions of lives each year. In the Indian subcontinent, the burden is particularly severe, with rising cases attributed to genetic predisposition, lifestyle factors, and limited access to early diagnostic tools. The increasing prevalence of CVD highlights the urgent need for reliable, efficient, and accurate diagnostic systems that can aid in early detection and intervention. In response to this need, the present study proposes an optimized hybrid deep learning model that combines the temporal learning capabilities of Recurrent Neural Networks (RNN) with the efficiency of Gated Recurrent Units (GRU). The objective is to enhance predictive performance for early diagnosis of heart disease. The model is trained and validated using a dataset of 918 patient records obtained from IEEE Dataport. To ensure data quality and robustness, extensive preprocessing steps were undertaken, including outlier removal through the Interquartile Range (IQR) method and normalization of numerical features to standardize the input space. Addressing class imbalance a common issue in medical datasets the Synthetic Minority Over-sampling Technique (SMOTE) was applied to generate a balanced distribution of positive and negative cases. The dataset was subsequently split into training and testing sets, allowing for comprehensive evaluation of the model. Hyperparameter tuning was conducted using GridSearchCV in conjunction with 10-fold cross-validation to optimize the model's performance. The hybrid RNN-GRU architecture significantly outperformed its individual counterparts in terms of predictive accuracy and generalization. The proposed model achieved outstanding results: 99.6% accuracy, 99.6% F1 score, 99.6% precision, and 99% recall. These metrics notably exceed those of existing models, which report accuracies ranging between 87% and 97%. The success of this hybrid approach lies in its ability to effectively capture and analyze temporal dependencies within the cardiovascular data. The RNN component facilitates sequential learning, while the GRU units mitigate issues related to vanishing gradients and enhance learning efficiency. Together, they enable the model to extract meaningful patterns from time-series data, such as heartbeat signals and patient history, that are critical for accurate diagnosis.