Utilization of machine learning for dengue case screening.

Area of Science:

• Public Health
• Artificial Intelligence
• Epidemiology

Background:

• Dengue infections cause significant global mortality and morbidity, necessitating improved control and diagnostic strategies.
• Artificial intelligence (AI), specifically machine learning (ML), offers potential for enhancing dengue management.
• Early and accurate screening of dengue cases is critical for effective public health interventions.

Purpose of the Study:

• To identify key variables for screening dengue cases using ML models.
• To evaluate the accuracy of different ML models in classifying dengue cases.
• To explore the feasibility of developing a mobile application for dengue screening.

Main Methods:

• Utilized data from reported dengue cases in Rio de Janeiro and Minas Gerais (2016, 2019) from the National Notifiable Diseases Surveillance System (SINAN).
• Employed the mutual information technique to determine the most relevant variables associated with confirmed dengue cases.
• Trained and tested ML models (Logistic Regression, Decision Tree, MLP) on a dataset of 10,000 confirmed and 10,000 discarded dengue cases, split into 70% training and 30% testing sets.

Main Results:

• Identified ten key variables for dengue screening: gender, age, fever, myalgia, headache, vomiting, nausea, back pain, rash, and retro-orbital pain.
• Logistic Regression, Decision Tree, and Multilayer Perceptron (MLP) models demonstrated high performance.
• Achieved a maximum accuracy of 98% in classifying dengue cases using the developed ML models.

Conclusions:

• Machine learning models, particularly tree-based approaches, are highly effective for accurate dengue case screening.
• The identified variables and high-accuracy models support the development of practical tools for healthcare professionals.
• A smartphone application utilizing a tree-based model could significantly aid in the rapid identification and management of dengue.