The West Nile virus (also known as West Nile Virus, WNV) infects thousands of people each year, causing symptoms ranging from severe fevers to severe neurological complications and even death in about 20% of cases.
In addition to humans, the virus also affects animals such as birds and horses, causing mortality rates of up to 40% in the latter. First isolated in 1937 in the West Nile district of Uganda, from which it takes its name, it has now spread worldwide.
For these reasons, the control and prevention of WNV infections is a topic of great interest.
Infected mosquitoes are the main vector for transmitting the virus to humans. When the first human cases were reported in Chicago in 2002, the city's Public Health Department initiated an extensive surveillance and control programme that remains in place today.
From late spring to early autumn, numerous mosquito traps are distributed throughout the area, testing each week for the presence of the virus in captured specimens. Based on this information, the city plans to use insecticides to control the adult mosquito population.
With an average of 91 active traps for 14 weeks each year, this programme presents significant costs for collecting samples and performing clinical tests to determine the presence of the virus.
This paper aims to design a supervised classification model capable of predicting whether or not a virus will be detected in a certain trap each week. The predictions of this model will allow collection efforts to be directed towards those traps where infected specimens are most likely to be found.
Given that tests over the years have only tested positive for the virus in 8.6% of cases, a sufficiently accurate model can greatly reduce the number of traps to be tested each week and, consequently, the costs of the program.