Building a one-vs-all classifier for spatial prediction of detected pathogens

More than 75% of human infectious diseases are caused by the transmission of pathogens from animals to humans (that is, zoonotic spillover). This demonstrates the importance of understanding the relative risks of each pathogen in each spatial region. In this study, we build one-vs-all classifiers to distinguish Mycobacterium and Listeria amongst all other recorded bacteria. We selected these two bacteria as they cause morbidity and fatality among humans and animals. We overcome the impact of class imbalance caused by spatial and taxonomical biases in detected pathogen occurrence data by under-sampling the majority negative samples and keeping all the minority positive samples. We further improved the prediction results by including animal richness data (number of genera present). Our findings highlight that there is a weak relationship between the predictive features and the relative occurrence of the target pathogen. We also identified that the inclusion of spatial-temporal information in the prediction process could increase generalisability. The biological study of the detected features suggests that more targeted infectious diseases surveillance data is required to validate the predicted results.

N/A

Proceedings of the International Congress on Modelling and Simulation, MODSIM

N/A

9780987214300

N/A

N/A

7