Student Research Reports

View Research Report

In recent years, mosquito-borne diseases such as the Zika virus, West Nile virus, Chikungunya virus, Dengue, and Malaria have become more prevalent in urban areas due to various climate and anthropogenic factors. This led to a greater need for mosquito abundance prediction to improve the response to disease outbreaks, especially during the summer when mosquito season peaks and outdoor activities increase significantly. The objective of this study was to evaluate the accuracy of six machine learning models for classifying extreme mosquito abundance events based on climate data. Data sourced from the Mosquito Habitat Mappers challenge on GLOBE and a City of Chicago dataset were matched to area-averaged time-series climate data for Chicago from GIOVANNI, a NASA open access remote sensing database for Earth science. Data was cleaned and then aggregated to a single weekly time-series dataset consisting of mosquito abundance, and the past week’s three climate variable averages. The models were trained and tested on climate data, namely surface humidity, precipitation, and daytime temperature. The mosquito and climate data were recorded from five Chicago summers. The results indicated that the best models for predicting mosquito abundance events were the ensemble learning methods of AdaBoost and Random Forest, respectively. Future avenues of research include using other, more-specific factors for prediction such as the chlorophyll from algal blooms (increasingly common due to direct and indirect anthropic activities, such as fertilizer runoff and warming waters due to climate change), more localized predictions, accounting for the microclimates of urban areas, and using regression models to predict precise mosquito numbers. Keywords: Mosquito Abundance, Machine Learning, Classification, Climate Variables, Remote Sensing