Machine-Learning-Based Forecasting of Dengue Fever in Brazilian Cities Using Epidemiologic and Meteorological...

Area of Science:

• Public Health
• Epidemiology
• Machine Learning

Background:

• Dengue poses a significant public health challenge globally and in Brazil.
• Current prevention strategies heavily rely on vector control and disease surveillance due to the lack of vaccines or specific treatments.
• Accurate and timely dengue forecasts are crucial for mitigating disease spread.

Purpose of the Study:

• To develop and evaluate a machine learning model for predicting monthly dengue cases in Brazilian cities one month ahead.
• To compare the performance of various machine learning algorithms and feature selection methods.
• To assess the potential of these models in enhancing dengue surveillance systems.

Main Methods:

• Utilized epidemiologic and meteorological data from 2007-2019 for model development.
• Compared machine learning algorithms including random forests, gradient boosting regression, feed-forward neural networks, and support vector regression.
• Employed feature selection methods to optimize model inputs for individual cities.

Main Results:

• The optimal model varied across different Brazilian cities.
• A random forests model trained on historical dengue cases demonstrated superior performance overall.
• The median prediction error across all cities was 12.2 cases, reduced to 11.9 when optimizing models per city.

Conclusions:

• Machine learning, particularly decision tree ensemble models, shows promise for improving dengue surveillance in Brazil.
• These models achieve low prediction errors, offering a valuable tool for public health interventions.
• Tailoring models and features to specific city characteristics can further enhance prediction accuracy.