Skip to main content
. 2018 Aug 2;18(8):2521. doi: 10.3390/s18082521

Figure 9.

Figure 9

Results for batch 27 processed by the GRU-autoencoder-powered anomaly detection, given its PSO-optimised parameters α (the start of the window for which an alert is for), β (the length of the window), and τ (the threshold which the loss between predicted values and actual values must cross to be considered an alert), in relation to the prevalence of respiratory disease within a single batch.