Dengue Severity Level Prediction Using Hematology-Driven Machine Learning Models

Dengue fever represents a significant public health challenge due to its potential to progress rapidly into severe and life-threatening conditions, emphasizing the need for early and reliable severity assessment. Using typically available complete platelet count numbers, this study proposes a hematology-driven machine learning architecture for triage-level dengue severity prediction. An informed severity variable based on clinical considerations was developed to classify the patients into minor, moderate, and severe according to platelet count, hematocrit, white blood cell counts, and hemoglobin based on a freely provided dataset of laboratory confirmed dengue cases. The dataset was obtained from a publicly available Kaggle repository and consists of laboratory-confirmed dengue patient records with complete hematological profiles, totaling 1,450 patient samples. Multi-class severity classification in the case of multi-class classification of severity, four ML algorithm, logistic regression, random forests, support vector machines, and extreme gradient boosting algorithms, trained machine learning models have been trained and evaluated .The extreme gradient boosting model achieved a precision of 64.1% and a macro-averaged F1-score of 0.59, outperforming logistic regression, random forests, and support vector machines by providing more balanced and consistent classification performance across minor, moderate, and severe dengue cases. Although the sensitivity maintained was clinically acceptable, in terms of class-wise analysis, there was a significant predictive ability of the minor and moderate severities. The most salient predictors of the severity were found to be white blood cell count, hematocrit, and platelet count in model interpretability analysis based on Shapley Additive Explanations. These results indicate that machine learning in combination with traditional hematological indicators will offer a viable, cost-efficient, and explainable method of early detecting dengue severity, thus offering meaningful clinical decision support in the medical care environment with scarce resources.