Limited Spectroscopy Data and Machine Learning for Detection of Zika Virus Infection in Aedes aegypti Mosquitoes

This study presents a technique for categorizing Aedes aegypti mosquitoes infected with the Zika virus under laboratory conditions. Our approach involves the utilization of the near-infrared spectroscopy technique and machine learning algorithms. The model developed utilizes the absorption of light from 350 to 1000 nm. It integrates Linear Discriminant Analysis (LDA) of the signal's windowed version to exploit non-linearities, along with Support Vector Machine (SVM) for classification purposes. Our proposed methodology can identify the presence of the Zika virus in intact mosquitoes with a balanced accuracy of 96% (row C2HT, average of columns TPR (%) and SPC (%)) when heads/thoraces of mosquitoes are scanned at 4, 7, and 10 days post virus infection. The model was 97.1% (10 DPI, row C2AB, column ACC (%)) accurate for mosquitoes that were used to test it, i.e., mosquitoes scanned 10-days post-infection and mosquitoes whose abdomens were scanned. Notable benefits include its cost-effectiveness and the capability for real-time predictions. This work also demonstrates the role played by different spectral wavelengths in predicting an infection in mosquitoes.