Fig 1.
Schematic diagram illustrating how early detection of norovirus was predicted using the classification of machine learning and a risk index.
L, M, and U denote the lower, average, and upper bounds of the confidence intervals of the norovirus warning, respectively.
Fig 2.
Weekly norovirus detection rate and meteorological characteristics curves (average of temperature, relative humidity, soil temperature at 3m, and day length) from 2009 to 2018 in South Korea.
Fig 3.
A) The F-statistic for 14 features computed from the univariate feature selections about weekly norovirus warning. B) Correlation analysis between the weekly norovirus warning and 11 selected features from the univariate feature selections.
Fig 4.
ROC curve comparing 6 machine learning algorithms for A) train and B) test data.
Table 1.
Performance of train and test data for predicting weekly norovirus warning, using the 6 machine learning algorithms.
Fig 5.
Comparison of early detection of norovirus between six machine learning algorithms for test data.
A) start week of early detection and B) end week of early detection. The vertical red dotted lines indicate an observed start and end week of norovirus warning phase. The horizontal color bars indicate a predicted interval for start and end week of norovirus warning phase using SVM (yellow), MLP (green), RF (pale blue), GB (blue), LSTM (orange), and GRU (red). The gray bars indicate an observed weekly detection rate of norovirus.
Table 2.
Comparison of observed start and end week of norovirus warning phase and predicted start and end week time interval of early detection of norovirus in 6 machine learning algorithms.
Fig 6.
Feature importance of A) RF and B) GB.