XAI-Driven Stacking Ensemble for Cross-Disease Feature Modification and Clinical Insights
Mehran University Research Journal of Engineering and Technology · 01.04.2026 · 45 (2) · 146-161
DOI: 10.22581/0651 ↗Kurz gefasst
Fachartikel in der Zeitschrift Mehran University Research Journal of Engineering and Technology, online erschienen am 01.04.2026. Verfasst von Shakeel Ahmad und 4 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
There is an increasing need to have timely and accurate clinical advice in management of diseases. Predictive models should be precise and in addition, clinically significant. The current literature is lacking in this aspect, and it does not satisfy this standard in many cases. This is filled by a complex also Stacking Classifier that attempts to predict in a multi-class manner five health conditions Bronchitis, Cold, Flu, Healthy and Pneumonia out of only structured symptom and vital sign data. The group incorporates the Random Forest (RF), Gradient Boosting (GB) and Support Vector Classifier (SVC) as the base learners, with a Logistic Regression (LR) meta-classifier condensing them. In contrast to the previous ones, which may rely only on high accuracy, our approach is based on the more developed interpretability framework, beginning with Explainable AI (XAI) through Local Interpretable Model-agnostic Explanations (LIME) to guarantee the transparency of predictions. The originality as well as its distinctiveness is the use of the Diverse Counterfactual Explanations (DiCE) framework that approximates the minimum changes in the independent features to convert a patient to a new disease and, thus, creates an individualized intervention approach. ADASYN pre-processing resulted in a good overall accuracy and Weighted F1-score of 94% in the model. Together with the actionability made available through DiCE, these overall results indicate that our proposed model is the most performant solution, which can provide excellent performance and unprecedented clinical value compared to current literature. This combined XAI framework extends past the conventional diagnostic tools to offer a proscriptive, action-based and trustworthy system of the personalized management of respiratory health. The performance rates of the proposed stacking model are also always higher compared to those presented in the literature of benchmarks on the case of multi-class respiratory disease prediction.
Zitieren
APA
Ahmad, S., Saqib, S. M., Syed, A. H., Abdelrahim, A. A., & Alromema, N. (2026). XAI-Driven Stacking Ensemble for Cross-Disease Feature Modification and Clinical Insights. Mehran University Research Journal of Engineering and Technology, 45(2), 146-161. https://doi.org/10.22581/0651
Vancouver
Ahmad S, Saqib SM, Syed AH, Abdelrahim AA, Alromema N. XAI-Driven Stacking Ensemble for Cross-Disease Feature Modification and Clinical Insights. Mehran University Research Journal of Engineering and Technology. 2026;45(2):146-161. doi:10.22581/0651
BibTeX
@article{Ahmad2026,
author = {Ahmad, Shakeel and Saqib, Sheikh Muhammad and Syed, Asif Hassan and Abdelrahim, Ali Ahmed and Alromema, Nashwan},
title = {XAI-Driven Stacking Ensemble for Cross-Disease Feature Modification and Clinical Insights},
journal = {Mehran University Research Journal of Engineering and Technology},
year = {2026},
volume = {45},
number = {2},
pages = {146--161},
doi = {10.22581/0651}
}
EUVIMED