Abstract
Heart failure (HF) detection is a critical task in medical diagnostics, often requiring accurate and efficient methods. This paper introduces a novel hybrid three stage expert system designed to improve HF detection. The proposed system integrates stacked autoencoder (AE) for feature extraction, an [Formula: see text]-penalized support vector machine (SVM) to select high quality subset of features from the autoencoded features, and a non-linear SVM for classification. The stacked AE extracts meaningful features from a set of HF risk factors, while the [Formula: see text]-SVM refines the feature set by selecting the most relevant features. In the final stage, a non-linear SVM with RBF kernel is used to classify the refined feature subset. The system is validated using a benchmark HF dataset, and its performance is evaluated using various metrics, including accuracy, sensitivity, specificity, Matthews correlation coefficient (MCC), and area under the curve (AUC). Experimental results show that the proposed system achieves an accuracy of 97.78%, sensitivity of 97.56%, specificity of 97.96%, and an MCC value of 0.955, outperforming current state-of-the-art methods. The system's ability to achieve high performance with a reduced feature set highlights its efficiency. The proposed approach provides a robust solution for HF detection, offering valuable decision support for healthcare professionals.