EBHOA-EMobileNetV2: a hybrid system based on efficient feature selection and classification for cardiovascular disease diagnosis.

Abstract

The accurate prediction of cardiovascular disease (CVD) or heart disease is an essential and challenging task to treat a patient efficiently before occurring a heart attack. Many deep learning and machine learning frameworks have been developed recently to predict cardiovascular disease in intelligent healthcare. However, a lack of data-recognized and appropriate prediction methodologies meant that most existing strategies failed to improve cardiovascular disease prediction accuracy. This paper presents an intelligent healthcare framework based on a deep learning model to detect cardiovascular heart disease, motivated by present issues. Initially, the proposed system compiles data on heart disease from multiple publicly accessible data sources. To improve the quality of the dataset, effective pre-processing techniques are used including (i) the interquartile range (IQR) method used to identify and eliminate outliers; (ii) the data standardization technique used to handle missing values; (iii) and the 'K-Means SMOTE' oversampling method is used to address the issue of class imbalance. Using the Enhanced Binary Grasshopper Optimization Algorithm (EBHOA), the dataset's appropriate features are chosen. Finally, the presence and absence of CVD are predicted using the Enhanced MobileNetV2 (EMobileNetV2) model. Training and evaluation of the proposed approach were conducted using the UCI Heart Disease and Framingham Heart Study datasets. We obtained excellent results by comparing the results with the most recent methods. The proposed approach beats the current approaches concerning performance evaluation metrics, according to experimental results. For the UCI Heart Disease dataset, the proposed research achieves a higher accuracy of 98.78%, precision of 99%, recall of 99% and F1 score of 99%. For the Framingham dataset, the proposed research achieves a higher accuracy of 99.39%, precision of 99.50%, recall of 99.50%, and F1 score of 99%. The proposed deep learning-based classification model combined with an effective feature selection technique yielded the best results. This innovative method has the potential to enhance the accuracy and consistency of heart disease prediction, which would be advantageous for clinical practice and patient care.