A hybrid stacking ensemble approach combining fuzzy logistic regression and fuzzy neural networks for enhanced COVID-19 diagnostic accuracy

Abstract

The timely and accurate diagnosis of COVID-19 remains crucial to managing the pandemic and preventing further transmission. Traditional diagnostic models often struggle with the complexity and uncertainty inherent in medical data related to COVID-19. This study proposes a novel hybrid stacking ensemble model that integrates fuzzy logistic regression (FLR) and fuzzy artificial neural networks (FANNs) based on Z-numbers. Z-numbers effectively capture uncertainty by representing both the restriction and reliability of medical information. By leveraging the complementary strengths of fuzzy logic and neural networks through a stacking ensemble technique, the proposed model enhances the robustness and accuracy of COVID-19 diagnostic predictions. The hybrid model was trained on a dataset of 200 patients, using 80% of the data for training and 20% for testing. The hybrid stacking ensemble model demonstrated an accuracy of 90.5% on a COVID-19 dataset, outperforming both fuzzy logistic regression (FLR) and fuzzy artificial neural networks (FANNs) individually. The area under the receiver operating characteristic curve (AUC-ROC) for the stacked model (0.93) is significantly higher than that of the individual models, confirming its superior diagnostic accuracy. Although the findings are based on a relatively small sample size (n = 200), which may limit generalizability and indicate the need for validation in larger and more diverse datasets, this innovative hybrid model offers a powerful tool for real-world medical diagnostics, providing significant implications for timely treatment and effective public health interventions in the fight against COVID-19 and potentially other infectious diseases.