AN EFFICIENT HEART DISEASE PREDICTION USING HYBRID DEEP LEARNING-BASED TWO-STAGE CLASSIFICATION MODEL

Heart disease is a fatal disease that affects people. It has the greatest world death rate of any type of disease. Thus, every preventive step must be done before the heart disease reaches its last stage. Thus, if we can scientifically identify cardiac disease at an early stage using different decision support systems without the use of any form of traditional methods, the worldwide death rate from this disease is certain to decrease. Through the development of highly advanced medical decision support systems, numerous researchers have studied the detection of heart disease. Artificial neural network concepts represent the highest predictive accuracy over medical data compared to other decision support systems. The primary goal of the proposed research is to use medical data and images to categorize and predict cardiac disease. A hybrid deep learning (DL)-based two-stage classification model for the prediction of heart disease is proposed in this research. In the first stage, data are gathered from the Cleveland UCI machine learning repository dataset for heart disease prediction; then, step two involves classifying echocardiography images to determine the heart disease. For medical data classification, a hybrid improved ResNet-34 (IResNet-34) with the MobileNetV2 technique is used. For echocardiogram image classification, a hybrid modified VGG-16 (MGG-16) with a DenseNet-169 model is used. Heart disease was predicted by combining and validating the classification results from both classification approaches. 96.76% accuracy in identifying normal data and 98.87% accuracy in identifying abnormal data were achieved using the MobileNetV2-IResNet-34 approach. The proposed MVGG-16 with the DenseNet-169 method gives better classification among echocardiogram images and it produces the maximum accuracy of 98.31%. The simulation findings show that, in terms of several performance matrices, the proposed hybrid DL approaches outperform the usual learning techniques for both medical data classification and medical image classification.