A hybrid optimization and tree-based learning framework for dengue diagnosis using hematological data

Dengue fever remains a major public health concern in tropical and subtropical regions, where rapid and accurate diagnosis is essential for timely medical intervention and disease control. This study introduces an explainable, nature-inspired machine learning framework for classifying dengue infection using routinely available hematological features obtained from complete blood count (CBC) analysis in the Bangladeshi context. The Grey Wolf Optimizer (GWO) was employed to fine-tune the hyperparameters of the Extra Trees Classifier (ETC), and its performance was compared with alternative tuning strategies under the same experimental protocol. The optimized GWO–ETC model achieved an accuracy of 95.78%, precision of 0.9541, recall of 0.9516, and F1-score of 0.9524, demonstrating strong internal test-set performance across the selected evaluation metrics. Explainable Artificial Intelligence (XAI) methods, including SHAP and LIME, were applied to interpret the decision-making process of the model. The analysis identified platelet count, plateletcrit (PCT), red cell distribution width (RDW-CV), red blood cell (RBC) count, lymphocytes, and monocytes as the most influential predictors of dengue infection. These findings are consistent with established clinical evidence linking platelet depletion and hematocrit variation with dengue pathophysiology. The proposed framework demonstrates interpretable classification capability, providing a transparent, data-driven approach that may assist clinicians in early dengue screening. By using accessible hematological parameters, the study supports cost-effective and explainable AI-driven decision support in resource-limited healthcare settings, while recognizing the need for further external validation and optimizer-level robustness analysis.