Enhancing Dengue Fever Diagnosis: A Machine Learning Framework with Stacking Ensemble and SHAP Explainability

Dengue fever remains a significant public health concern in tropical and subtropical regions, necessitating rapid and accurate diagnosis to enhance patient management and resource allocation. This study presents a machine learning framework leveraging hematological data for effective dengue classification. The dataset, comprising 17 clinical and demographic features, was meticulously preprocessed through outlier removal, missing value imputation, label encoding of categorical variables, and feature normalization using Standard Scaler. A total of 11 established machine learning algorithms were evaluated using GridSearchCV and 5-fold crossvalidation. Among them, a fine-tuned Stacking Ensemble model demonstrated superior performance, achieving 77% accuracy on the original dataset and 83% accuracy after applying SMOTE-based oversampling. In contrast, Support Vector Machine outperformed others under under sampling conditions with 66 % accuracy. Feature importance and model transparency were ensured via Shapley Additive Explanations (SHAP), offering interpretability into critical diagnostic features. These findings emphasize the potential of ensemble models augmented with explainability tools in clinical decision support. Future work aims to expand the model's scope to incorporate Banglish-language clinical texts and employ advanced deep learning architectures such as LSTM and Transformers for enhanced diagnostic precision.