Dengue fever presents severe diagnostic challenges in tropical countries where specialized antigen or antibody testing kits are frequently delayed or unavailable in rural clinics. Routine Complete Blood Count (CBC) tests, however, are ubiquitous and fast.
Our research investigated how standard CBC parameters can be used with supervised machine learning algorithms for reliable early dengue detection.
- Research and Project Details: https://kholipha-ahmmad-al-amin.me/#research
- Author Portfolio: https://kholipha-ahmmad-al-amin.me/
- Machine Learning Projects: https://kholipha-ahmmad-al-amin.me/#projects
Methodology and Findings
- CBC Feature Selection: Evaluated key blood parameters including platelet count, white blood cell count (WBC), hematocrit, lymphocyte percentage, and neutrophil ratios.
- Model Benchmark: Compared multiple classifiers including Random Forest, Support Vector Machines (SVM), and Gradient Boosting (XGBoost).
- Interpretability with SHAP: Used Shapley Additive Explanations to verify that feature importance aligned with clinical hematological literature, ensuring models did not rely on spurious correlations.
- Cross-Validation: Rigorous 10-fold stratified cross-validation ensured models maintained high sensitivity, minimizing false-negative diagnostic classifications.
Read the full research abstract, methodology, and notebook repositories at https://kholipha-ahmmad-al-amin.me/#research.
Top comments (0)