Using Machine Learning Regressors for the Discovery of Culex Mosquito Habitats and Breeding Patterns in Washington D.C.

Abstract

Culex mosquitoes pose a significant threat to humans and other species due to their ability to carry deadly viruses such as the West Nile and Zika. Washington D.C., in particular, has a humid subtropical climate that is ideal as a habitat for mosquito breeding. Thus, tracking mosquitoes’ habitats and breeding patterns in Washington D.C. is crucial for addressing local public health concerns. Although fieldwork techniques have improved over the years, monitoring and analyzing mosquitoes is difficult, dangerous, and time-consuming. In this work, we propose a solution by creating a Culex mosquito abundance predictor using machine learning techniques to determine under which conditions Culex mosquitoes thrive and reproduce. We used four environmental variables to conduct this experiment: precipitation, specific humidity, enhanced vegetation index (EVI), and surface skin temperature. We obtained sample data of these variables in the Washington D.C. areas from the NASA Giovanni Earth Science Data system, as well as mosquito abundance data collected by the D.C. government. Using these data, we created and compared four machine learning regression models: Random Forest, Decision Tree, Support Vector Machine, and Multi-Layer Perceptron. We searched for the optimal configurations for each model to get the best fitting possible. Random Forest Regressor produced the most accurate prediction of mosquito abundance in an area with the four environment variables, achieving a mean average error of 3.3. EVI was the most significant factor in determining mosquito abundance. Models and findings from this research can be utilized by public health programs for mosquito-related disease observations and predictions.