Machine Learning-Based Prediction of Dengue Infection Using Clinical and Hematological Data from Bangladesh: A Comparative Analysis of Benchmark and Extended Prediction Models

Neesha Naimeen; Victor Miranda Soberanis

cs.LGOct 10, 2026

Machine Learning-Based Prediction of Dengue Infection Using Clinical and Hematological Data from Bangladesh: A Comparative Analysis of Benchmark and Extended Prediction Models

Authors: Neesha Naimeen, Victor Miranda Soberanis

Organizations: School of Engineering, Computer and Mathematical Sciences, Auckland University of Technology (AUT), Auckland,New Zealand · Department of Mathematical Sciences, Auckland University of Technology (AUT), Auckland,New Zealand

Abstract

Dengue is a viral infection caused by the dengue virus (DENV) and a major public health concern in Bangladesh, where symptom overlap with other febrile illnesses complicates early diagnosis. This study compares Logistic Regression, Random Forest and XGBoost to predict dengue from clinical and hematological data using two Bangladeshi hospital datasets (2024, n=1,511; 2025, n=1,283) with SMOTE. Extended models outperformed benchmarks within 2024 (Random Forest ROC-AUC 0.640 to 0.661), with RDW-CV the most consistently influential added predictor. 2025 benchmark performance was higher (ROC-AUC 0.904), unexplained by class imbalance and possibly reflecting data quality. External validation is needed.

Figures & tables

Variable	Dataset 1 (2024)	Dataset 2 (2025)	Notes
Age	Age	Age	Common; D1 integer years; D2 has a 0.5 value (infant, in years)
Sex	Gender	Sex	Common; Male/Female, no missing
Red blood cell count	RBC	RBC	Common; Same scale (millions/µL) but D2 has extreme outliers (min 0.5, max 66.9) needing cleaning.
Haematocrit	HCT(%)	HCT	Common; Both (%),comparable range
Neutrophils %	Neutrophils(%)	Neut%	Common
Lymphocytes %	Lymphocytes(%)	Lymph%	Common

Table 1: Predictor correspondence between Dataset 1 (2024) and Dataset 2 (2025)

	Recall (Negative)			F1-score			ROC-AUC
Model	None	SMOTE	Under	None	SMOTE	Under	None	SMOTE	Under
Logistic Regression	0.189	0.484	0.453	0.825	0.679	0.646	0.570	0.570	0.552
Random Forest	0.274	0.347	0.463	0.834	0.816	0.711	0.635	0.640	0.610
XGBoost	0.347	0.379	0.526	0.792	0.786	0.632	0.647	0.645	0.588

Table 2: 2024 benchmark model: effect of class-imbalance handling

Figure 1: Confusion matrices – 2024 Benchmark Model (8 common predictors, SMOTE)

Figure 2: ROC curves – 2024 Benchmark Model (8 common predictors, SMOTE)

Figure 3: Feature importance for the 2024 benchmark model.

Model	Accuracy	Precision	Recall	F1-score	ROC-AUC	Recall (Neg)	Precision (Neg)
Logistic Regression	0.587	0.729	0.635	0.679	0.570	0.484	0.377
Random Forest	0.723	0.750	0.894	0.816	0.640	0.347	0.600
XGBoost	0.690	0.746	0.832	0.786	0.645	0.379	0.507

Table 3: 2024 Benchmark Model performance (8 common predictors, SMOTE-balanced)

Figure 4: Confusion matrices – 2024 Extended Model (18 predictors, SMOTE)

Figure 5: ROC curves – 2024 Extended Model (18 predictors, SMOTE)

Figure 6: Random Forest feature importance – 2024 Extended Model (18 predictors, SMOTE)

Model	Accuracy	Precision	Recall	F1-score	ROC-AUC	Recall (Neg)	Precision (Neg)
Logistic Regression	0.653	0.754	0.736	0.745	0.635	0.474	0.450
Random Forest	0.762	0.770	0.933	0.843	0.661	0.389	0.725
XGBoost	0.726	0.757	0.885	0.816	0.664	0.379	0.600

Table 4: 2024 Extended Model performance (18 predictors, SMOTE-balanced)

Model	F1-score		$\Delta$	ROC-AUC		$\Delta$
	Benchmark	Extended	F1	Benchmark	Extended	AUC
Logistic Regression	0.679	0.745	+0.066	0.570	0.635	+0.065
Random Forest	0.816	0.843	+0.027	0.640	0.661	+0.021
XGBoost	0.786	0.816	+0.030	0.645	0.664	+0.019

Table 5: 2024: Benchmark (8 predictors) vs Extended (18 predictors), identical rows/split

	Recall (Negative)			F1-score			ROC-AUC
Model	None	SMOTE	Under	None	SMOTE	Under	None	SMOTE	Under
Logistic Regression	0.446	0.785	0.769	0.891	0.874	0.887	0.844	0.848	0.857
Random Forest	0.631	0.723	0.785	0.920	0.915	0.895	0.895	0.904	0.905
XGBoost	0.692	0.677	0.769	0.929	0.919	0.860	0.903	0.902	0.892

Table 6: 2025 benchmark model: effect of class-imbalance handling

Figure 7: Confusion matrices – 2025 Benchmark Model (8 common predictors, SMOTE)

Figure 8: ROC curves – 2025 Benchmark Model (8 common predictors, SMOTE)

Model	Accuracy	Precision	Recall	F1-score	ROC-AUC	Recall (Neg)	Precision (Neg)
Logistic Regression	0.821	0.920	0.833	0.874	0.848	0.785	0.614
Random Forest	0.872	0.908	0.922	0.915	0.904	0.723	0.758
XGBoost	0.875	0.896	0.943	0.919	0.902	0.677	0.800

Table 7: 2025 Benchmark Model performance (8 common predictors, SMOTE-balanced)

Model	2024 F1	2025 F1	2024 ROC-AUC	2025 ROC-AUC
Logistic Regression	0.679	0.874	0.570	0.848
Random Forest	0.816	0.915	0.640	0.904
XGBoost	0.786	0.919	0.645	0.902

Table 8: Benchmark model comparison: 2024 vs 2025 (identical 8 predictors)

Predictor	Random Forest	XGBoost	LR $\|$ coef $\|$	Mean rank
Monocytes%	0.0587	0.1043	0.3012	1.67
Eosinophils%	0.0546	0.1283	0.2005	2.33
RDW-CV	0.0835	0.0672	0.1585	3.00
MCH	0.0545	0.0419	0.1853	4.00
PCT	0.0566	0.0386	0.1516	4.67
MPV	0.0494	0.0317	0.1345	7.00

Table 9: Extended predictors ranked by mean importance rank across three models

Predictor	2024 individual AUC	2025 individual AUC
Age	0.553	0.564
RBC	0.583	0.531
HCT	0.544	0.546
Neutrophils%	0.571	0.555
Lymphocytes%	0.580	0.560
Platelet Count	0.528	0.870

Table 10: Individual predictor discriminative power (univariate ROC-AUC against Result), by year

Predictor	2024 distinct values ( $n=1{,}511$ )	2025 distinct values ( $n=1{,}283$ )	Coarser in
Age	67	59	2025
RBC	4	480	2024
HCT	829	392	2025
Neutrophils%	23	183	2024
Lymphocytes%	23	172	2024
Platelet Count	1,157	332	2025

Table 11: Measurement resolution: number of distinct values per predictor, by year

Uncertainty	Question raised	Proposed design
Cause of the 2024–2025 gap	Does it reflect data quality, case mix or recording practice?	Data audit with the source hospitals; compare patient characteristics and laboratory methods
Transferability	Does a model trained on one dataset work on the other?	Cross-dataset training and testing on the eight common predictors
Stability of estimates	How much do results depend on one split?	Repeated stratified cross-validation with confidence intervals
Role of RDW-CV	Does it add value beyond 2024?	Independent validation in a dataset recording RDW-CV

Table 12: Open uncertainties and proposed follow-up designs

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Dataset	$n$	Positive (%)	Negative (%)
2024 Benchmark (8 predictors)	1,511	68.5	31.5
2024 Extended (18 predictors)	1,511	68.5	31.5
2025 Benchmark (8 predictors)	1,283	74.8	25.2

Appendix

Table 13: Dataset sizes and class balance after preprocessing

Task	Library
Data handling and cleaning	pandas, numpy
Data visualisation	matplotlib, seaborn
Machine learning models	scikit-learn
XGBoost model	xgboost
Class imbalance handling	imbalanced-learn (SMOTE, RandomUnderSampler)
Model evaluation	sklearn.metrics

Appendix

Table 14: Software and tools used for data processing, modelling, and evaluation

Explore similar work

TREA-Net: A Transferable Residual Epidemiological Adaptation Network for Dengue Incidence Forecasting
Jul 29, 2026Inesh Shukla, Madhurima Panja, Tanujit Chakraborty +1Epidemic Forecasting Residual Learning
International Institute of Information Technology Hyderabad, India · SAFIR, Sorbonne University Abu Dhabi, United Arab Emirates · Sorbonne Center for Artificial Intelligence, Sorbonne Universit´e, Paris, France
A Logistic Regression Model to Predict Malaria Severity in Children
May 17, 2026Mary Opokua Ansong, Asare Yaw Obeng, Samuel King OpokuLogistic Regression Clinical Outcome Prediction
Machine Learning-Based Prediction of Childhood Stunting in Bangladesh: Fairness and Temporal Robustness Assessment
Sep 21, 2026Md Ahshanul Haque, Muhammad Ashad KabirHealthcare Algorithmic Fairness
School of Computing, Mathematics and Engineering, Charles Sturt University, NSW, Australia

Jul 29, 2026J/K move · Enter open · S save

TREA-Net: A Transferable Residual Epidemiological Adaptation Network for Dengue Incidence Forecasting

Inesh Shukla, Madhurima Panja, Tanujit Chakraborty, Chittaranjan Hens

International Institute of Information Technology Hyderabad, India · SAFIR, Sorbonne University Abu Dhabi, United Arab Emirates · Sorbonne Center for Artificial Intelligence, Sorbonne Universit´e, Paris, France

Epidemic Forecasting Residual Learning Time Series Forecasting

Open paper AskPDF arXiv

Accurate multi-week dengue forecasting supports timely vector-control interventions, outbreak preparedness, and healthcare resource allocation. However, newly established surveillance systems often lack the historical data needed to train reliable neural forecasting models. Although pretrained time-series models offer promising zero-shot forecasts, their cross-domain training may not capture local epidemiological dynamics. We propose TREA-Net, a Transferable Residual Epidemiological Adaptation Network for dengue forecasting under limited data. TREA-Net augments neural forecasting backbones with projections from an Environmental Time-Series Susceptible-Infected-Recovered model and learns a lightweight gated residual correction transferable from data-rich to data-scarce regions. Its node-invariant design accommodates surveillance systems with different numbers of locations, while target adaptation requires learning only two global parameters. We transfer knowledge from long-running dengue surveillance in Colombia and Nicaragua to 8-week-ahead forecasting in Mexico and Malaysia using only 78 or 104 weeks of target data. Across five neural backbones and ten transfer settings, TREA-Net improves the corresponding backbone in 9 out of 10 settings, with statistically significant gains. When integrated with TiRex, a foundation model for forecasting, it achieves the lowest mean absolute error across all target datasets. Conformal prediction further maintains empirical coverage while reducing 8-week prediction-interval width by 29.6% in Mexico. These results demonstrate TREA-Net's potential as a lightweight and portable early-warning framework for health agencies with limited surveillance data.

Abstract

Explore similar work

TREA-Net: A Transferable Residual Epidemiological Adaptation Network for Dengue Incidence Forecasting

A Logistic Regression Model to Predict Malaria Severity in Children

Machine Learning-Based Prediction of Childhood Stunting in Bangladesh: Fairness and Temporal Robustness Assessment