Fluorescence excitation-emission matrix spectroscopy combined with machine learning for the classification of viruses for respiratory infections.

Significant efforts were currently being made worldwide to develop a tool capable of distinguishing between various harmful viruses through simple analysis. In this study, we utilized fluorescence excitation-emission matrix (EEM) spectroscopy as a rapid and specific tool with high sensitivity, employing a straightforward methodological approach to identify spectral differences between samples of respiratory infection viruses. To achieve this goal, the fluorescence EEM spectral data from eight virus samples was divided into training and test sets, which were then analyzed using random forest and support vector machine classification models. We proposed a novel strategy for data fusion based on fast Fourier transform (FFT) and wavelet transform (WT) methods, which significantly enhanced classification accuracy from 45 % to 75 %. This approach improved the classification capability for similar spectral characteristics of viruses. Rhinovirus was further differentiated from rotavirus, while influenza A virus was distinguished from inactivated poliovirus vaccines and rhinovirus. This study demonstrated that the integration of fluorescence EEM spectroscopy with machine learning algorithms presented significant potential for the detection of unidentified harmful substances in the ambient environment.