Accurate Heart Disease Prediction Using an Optimized InceptionSqueeze-and-Excitation Convolutional Neural Network

Introduction: Cardiovascular diseases affect the heart and blood vessels, and there is a need to diagnose these conditions at an early stage in order to increase survival rates considerably. Recent breakthroughs in Artificial Intelligence (AI) have created new opportunities for accurate heart disease prediction. However, many of the existing algorithms can still struggle with matters of accuracy, from bias to overfitting. In this research, a novel approach by which deep learning is exploited to enhance heart disease prediction with the aid of indicative key features is presented. The Shapley additive explanations algorithm can emphasize important features and compensate for the disadvantages of other black-box approaches, since it exposes the causal relationships between feature variables. Methods: The method starts from the data preprocessing step to make the quality of the dataset better by removing outliers, imputing missing values, and handling the problem of class imbalance. Then, discrete wavelet transforms and deep learning models are employed to capture handcrafted and deep features simultaneously. Afterwards, an enhanced Parrot optimization algorithm is used to perform feature selection. Results: The proposed method outperforms state-of-the-art methods with corresponding accuracies of 99.84% (Cleveland), 99.27% (Framingham Heart Study), and 99.73% (Statlog). Discussion: The Multi-Output generalized stacking with Inception Squeeze-and-Excitation Convolutional Neural Network (MOISECNN) is employed for classification, and the adaptive pied kingfisher optimizer algorithm is used to optimize the parameters of MOISECNN. Conclusion: The framework offers a global explanation analysis through the Shapley additive explanations algorithm, which identifies important attributes as well as alleviates the limitations of some existing black-box methodologies to explain machine learning models by explaining causations between features.