Fig. 5

Variable importance analysis using a random forests machine learning algorithm to determine the relative importance of symptom and demographic variables as determinants of the outcome of SARS-CoV-2 PCR testing, using 1125 cases and 4750 non-cases presenting for testing during the first wave of COVID-19 in New Zealand. The mean decrease in accuracy measures the importance of removing each variable on the predictive accuracy of the model, whereas the mean decrease in purity measures the importance of removing each variable on node ‘purity’ (see Methods)