Detection of Zika virus infection on mosquitoes using spectroscopy and machine learning

This work shows a method to classify mosquitoes infected with the Zika virus. We accomplish that by using spectroscopy and machine learning. Our model takes the light absorbance of wavelengths from 350 to 1000 nm as inputs. It employs a combination of Linear Discriminant Analysis (LDA) of the windowed version of the signal (to take advantage of nonlinearities) and Support Vectors Machine (SVM) to classify the samples. The proposed method can detect the presence of the Zika virus with 100% accuracy in less than 7 days post-infection. The accuracy drops to 77.7% when 10 days have passed. The main advantages are the low cost and the possibility to make predictions in real-time.