Next, the average retrospective testing accuracies of the BP neural
network and SVM were compared. Table 2 shows that both accuracies
were the least in the 60%–70% interval and the most in the 90%–100%
interval. However, the number of pentads of SVM in the 70%–80% interval
and the 90%–100% interval was more than the BP neural network. It
follows that the back detection accuracy rate of SVM is higher than for
the BP neural network, and the SVM model is more stable and superior
to the BP neural network model in actual prediction.