Modeling Dengue vector population using remotely sensed data and machine learning.

Area of Science:

• Environmental Science
• Public Health
• Remote Sensing

Background:

• Aedes aegypti mosquitoes transmit diseases like Dengue and Zika globally.
• Traditional vector surveillance methods are costly and time-consuming.
• Remote sensing offers a cost-effective alternative for monitoring mosquito populations.

Purpose of the Study:

• To develop and evaluate machine learning models for temporal prediction of Aedes aegypti oviposition activity.
• To compare the performance of machine learning models against linear models using satellite-derived environmental data.
• To provide improved tools for operational vector surveillance systems.

Main Methods:

• Time series analysis of Aedes aegypti oviposition data from ovitraps (2012-2016).
• Utilized satellite-derived variables: NDVI, NDWI, LST (night and day), and TRMM-GPM rainfall.
• Assessed machine learning models: Support Vector Machine, Artificial Neural Networks, K-Nearest Neighbors, and Decision Tree Regressor, alongside linear models.

Main Results:

• Machine learning models outperformed traditional linear models in predicting oviposition activity.
• K-Nearest Neighbors Regression (KNNR) demonstrated the highest predictive accuracy.
• The study validated model performance through parameter tuning and robust training/validation approaches.

Conclusions:

• Machine learning techniques provide superior, non-parametric alternatives for modeling nonlinear relationships in vector ecology.
• The developed models offer enhanced capabilities for operational implementation in public health surveillance systems.
• Accurate, timely prediction of Aedes aegypti activity can significantly improve disease vector control strategies.