Abstract
Heart disease continues to rank among the world's top causes of death, underscoring the pressing need for precise and accurate prediction techniques. Performance issues with traditional machine learning techniques have been identified, particularly when working with high-dimensional and unbalanced medical datasets. We present a new deep learning-based framework called IDLHICNet, or an Improved Deep Learning-based Hybrid Inception-Capsule Network, which is combined with an Enhanced Whale Optimization Algorithm (EWOA) for feature selection in order to overcome this issue. Using Improved K-Means Clustering (IKC) to remove outliers, Min-Max normalization to scale features, SMOTE oversampling to balance classes, and EWOA to select essential features are some of the crucial steps in the proposed process. The IDLHICNet model, which makes use of Capsule Networks' spatial awareness and the Inception architecture's feature extraction capabilities, is then used to classify the processed information. The effectiveness of our approach is demonstrated by experiments carried out on three benchmark datasets, including the Faisalabad, CVD, and heart failure datasets. The proposed model surpasses existing state-of-the-art methods by achieving high performance metrics, with accuracy values of 99.51%, 98.76%, and 99.07% across different test scenarios, as well as superior precision, recall, and F1-scores. This research demonstrates how well a hybrid deep learning architecture and sophisticated feature selection work together to predict heart disease accurately and early, allowing for prompt medical intervention and better patient outcomes.