Modified bi-directional long short-term memory and hyperparameter tuning of supervised machine learning models for cardiovascular heart disease prediction in mobile cloud environment

In recent decades, cardiovascular heart disease (CHD) has emerged as the predominant cause of mortality on a global scale. Numerous risk factors for cardiovascular disease necessitate prompt access to accurate, dependable, and efficient early diagnosis and disease management strategies. The timely and precise identification of cardiac disease holds significant importance in the prevention and treatment of heart failure. The reliability of diagnosing heart disease by standard medical history has been widely questioned. Medical imaging, including angiography, is the most widely used method for diagnosing CAD. Nevertheless, angiography is renowned for its high cost and its correlation with many adverse consequences. Noninvasive techniques, such as machine learning, demonstrate reliability and efficiency in the classification of individuals as either healthy or afflicted with cardiac disease. In this research, a new efficient real-time diagnostic method for CHD was designed via hyperparameter tuning of supervised machine learning (SML). This study consists of five significant components: the data preprocessing stage, classification with and without hyperparameter tuning, model integration and clinical decision, intervention and patient engagement, and lastly, model deployment and API integration on Amazon Cloud SageMaker. The data was preprocessed, which included finding outliers using the interquartile range (IQR), normalizing the data, and using the synthetic minority oversampling method (SMOTE) to fix the imbalance in the data. Following this, we propose a new adaptation of the Bi-Directional Long-Short Term Memory (BiLSTM) model, incorporating an attention mechanism (AM) to facilitate feature selection. The feature relevance was determined by calculating the coefficient scores for the top ten (10) features in each of the SMLs, namely the Catboost, extreme gradient boosting (XgBoost), random forest (RF), and linear discriminant analysis (LDA). Both with and without the inclusion of hyperparameter tuning, the grid search process further improved the models. We deployed our proposed models at scale by outsourcing them to the SageMaker cloud environment. The detailed experimental analysis showed that the BiLSTM + AM-XgBoost model worked best when tuned for hyperparameters via the grid search method. Specifically, this model achieved an accuracy of 99.99%, recall of 99.65%, precision of 99.68%, and AUC of 88.16%. The BiLSTM + AM-XgBoost model, without any hyperparameter adjustment, achieved an accuracy of 97.58%, recall of 96.32%, precision of 97.32%, and AUC of 87.32%. The tuning of hyperparameters affects the performance of SML in terms of accuracy, recall, precision, and AUC. In addition, we conducted a comparative analysis of our proposed models against the current state-of-the-art approaches. The findings of our evaluation indicate that our models exhibit a level of performance that outperforms the existing methods. The proposed model can function as a decision support system that employs SML techniques and seeks to enhance the efficacy of cardiac patient diagnostics for healthcare professionals. The simplicity of the suggested model is attributed to its reduced computational complexity. The findings demonstrate the potential application of our method in the field of telemedicine, particularly in the context of mobile cloud device environments for CHD.