Figure 11
Two multi-line graphs are presented side by side. In both graphs, the vertical axis is labeled “Mean Squared Error (M S E),” ranging from 10 to the negative 3 power to 10 to the 1 power in increments of 10 to the 1 power. The horizontal axis is labeled “Epoch,” ranging from 5 to 30 in increments of 5 in graph (a) and ranging from 5 to 40 with an interval of 5 in graph (b). In both graphs, four curves are drawn, representing “Train,” “Validation,” “Test,” and “Best.” On the left, the graph (a) is titled “Best validation performance is 0.0028096 at epoch 27.” The train, validation, and test curves start from around 10 to the 0 power on the vertical axis, and these decrease rapidly and initially overlap and then spread at the bend below the error of 10 to the negative 2 power just before the epoch of 10. After that, these curves remain constant and end on the right side around the error of 3.5 times 10 to the negative 3 power. The best curves are represented by the horizontal and vertical lines intersecting around the error of 2.9 times 10 to the negative 3 power at the epoch of 27. On the right, the graph (b) is labeled “Best validation performance is 0.0031696 at epoch 38.” The train, validation, and test curves start just above 10 to the 0 power on the vertical axis, and these decrease gradually in a concave-up pattern with slight fluctuation. The test curve remains at the top, which ends around 6.3 times 10 to the negative 3 power, while the train and validation curves end around 3.1 times 10 to the negative 3 power. The best curves are represented by the horizontal and vertical lines intersecting around the error of 3.1 times 10 to the negative 3 power at the epoch of 38. Note: All numerical data values are approximated.MSE curves over epoch: (a) LM network with 10-neuron hidden layer and (b) SCG network with 12-neuron hidden layer