We also obtain loss and mae values for this new model, but again, the numbers obtained are not directly comparable to the earlier two models for the log scale:

# Model evaluation
model %>%  evaluate(test, testtarget) 

OUTPUT
## $loss
 ## [1] 0.02701566
 ##
 ## $mean_absolute_error
 ## [1] 0.1194756 

pred <- model %>%  predict(test)
 plot(testtarget, pred)

We obtain a scatter plot of the actual values (log transformed) versus the predicted values based on the test data. We also get a scatter plot of the actual versus predicted values in the original scale for comparison with earlier plots. The scatter plots for predicted versus actual response values (model_three) are as shown in the following graph:

From the preceding graph, we can see that the significant underestimation pattern observed in earlier models shows improvement, both in the log scale and in the original scale. In the original scale, the data points at higher values are relatively closer to the diagonal line, indicating improved prediction performance by the model.