Cardiac Heart Disease Prediction using Deep Learning-based Ensemble Classifier

Abstract

Cardiac heart disease (CHD) is the most fatal cause of death globally. Early detection of CHD helps to take preventive action to improve the lifestyle of the patient. This paper presents a hybrid ensemble classifier for the CHD prediction based on the Deep Convolution Neural Network (DCNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) layer. The DCNN enhances the spatial characteristics, LSTM provides long-term dependency, and GRU depicts the temporal characteristics of the patient's healthcare attributes. Further, improved Synthetic Minority Oversampling Techniques (ISMOTE) are used for data augmentation to minimize the class imbalance problem and achieve stability in training. The effectiveness of the suggested system is evaluated on the Framingham dataset consisting of The EC provides a recall of 98.7%, a precision of 99.5%, a selectivity of 97.3%, an F1-score of 99.1%, a negative predictive value (NPV), and an accuracy of 98.5% for the original dataset. The EC-ISMOTE offers the improved recall of 99.8%, precision of 99.8%, selectivity of 99.8%, NPV of 99.8%, F1-score of 99.8% and accuracy of 99.8% which has shown a noteworthy boost over the performance of traditional CHD techniques.