Fig 1.
Graphical illustration of data processing procedure.
Data were obtained from records of swine submissions supplied from farms in Ontario, Canada, to the Animal Health Laboratory over the period from May 2007 to December 2015.
Fig 2.
Number of diagnostic submissions and positive virological submissions for IAV per week and month.
The counts were obtained from swine samples submitted to the Animal Health Laboratory in Ontario from May 2007 (week 19) to December 2015 (week 52). The original data are represented by blue lines for diagnostic submissions and by pink lines for positive virological submissions. The four time series were subjected to the decomposition, and the somewhat upward trend-cycle component in diagnostic submissions is shown in black while the slow increasing trend in positive counts is displayed in red.
Fig 3.
Retrospective predicted counts of weekly and monthly submissions and positive submissions for IAV.
The autoregressive integrated moving average (ARIMA) is shown in red, the generalized linear autoregressive moving average (GLARMA) in blue, and the random forest (RF) in green. The actual observations are represented by black lines.
Table 1.
Predictive accuracy evaluated via the root mean square error (RMSE) of autoregressive integrated moving average (ARIMA), generalized linear autoregressive moving average (GLARMA), and random forest (RF) time series models.
Table 2.
Retrospective random forest variable importance measurements.
Fig 4.
Prospective simulated counts of weekly and monthly submissions and positive submissions for IAV.
Counts were predicted for the last three years. The autoregressive integrated moving average (ARIMA) is shown in red, the generalized linear autoregressive moving average (GLARMA) in blue, and the random forest (RF) in green. The actual observations are represented in black.
Fig 5.
The residuals were obtained after fitting with simulated prospective autoregressive integrated moving average (ARIMA), generalized linear autoregressive moving average (GLARMA), and random forest (RF) model predicted counts at weekly and monthly intervals.
Table 3.
Confusion matrix for predicted monthly positive submissions with the prospective autoregressive integrated moving average (ARIMA), generalized linear autoregressive moving average (GLARMA) and random forest (RF) time series models.
Fig 6.
Normal Quantile-|Quantile (Q-Q) plots of the residuals.
Residuals were obtained after fitting with simulated prospective autoregressive integrated moving average (ARIMA), generalized linear autoregressive moving average (GLARMA), and random forest (RF) model predicted counts at weekly and monthly intervals.