Figure 3.

Plot of the feature importance across 100 replicates showing the features’ ranking for the prediction of udder health status of dairy cows according to somatic cell count below or above 200,000 cells/mL. Evaluated features, using neural network as predictive methods, are: log-transformed somatic cell count (log₁₀SCC), stage of lactation (DIM), differential somatic cell count (DSCC), protein, parity order, casein, log₁₀SCC of cows sampled in the same herd and day (log₁₀SCC_HTD), lactose, fat, season of sampling, milk yield (MY), year of sampling, pH, MY of cows sampled in the same herd and day (MY_HTD) and urea.