Temporal Machine Learning Models for Classifying Suspected Dengue Cases in Mexico Using Surveillance Data from 2025.

Area of Science:

• Epidemiology
• Machine Learning
• Public Health

Background:

• Dengue fever presents a significant public health challenge in Mexico, characterized by seasonal patterns and geographic variations.
• Recent epidemiological data are crucial for enhancing predictive models and reflecting current disease dynamics.

Purpose of the Study:

• To develop and compare temporal machine learning models for classifying dengue cases in Mexico using 2025 surveillance data.
• To evaluate the performance of nine different machine learning algorithms for binary classification of confirmed versus negative dengue cases.

Main Methods:

• Analysis of 68,222 suspected dengue cases from 2025 national surveillance data.
• Chronological data splitting into training, validation, and testing sets.
• Evaluation of nine machine learning algorithms including Random Forest, Bayesian Network, XGBoost, and others, with performance metrics like accuracy, F1-score, and AUC.

Main Results:

• Random Forest exhibited the best overall performance, achieving the highest F1-score (0.7254) and PR-AUC (0.7300).
• Bayesian Network yielded the highest accuracy (0.7023) and ROC-AUC (0.7756), but with less operational balance.
• Geographic and institutional factors were identified as key predictors, while comorbidities had less influence.

Conclusions:

• Temporal machine learning models are valuable tools for dengue case classification in Mexico.
• The Random Forest model offers a robust and balanced approach for dengue surveillance, crucial for public health.
• Accurate identification of confirmed dengue cases is vital for effective public health interventions.