In the modeling process of Artificial Neural Network (ANN)

an appropriate partition of train samples and test samples conduces to a high modeling efficiency.In most cases

test samples are selected randomly.In this paper

a Euclidian distance based self-clustering algorithm is proposed to partition train samples and test samples automatically.All the samples are attributed to different clusters firstly; each individual sample is affiliated to the nearest cluster according to the Euclidian distance from the kernels of other clusters whereafter; and then small clusters are divided up into others;at last test samples are selected from each cluster in the same proportion.Two case studies show that this algorithm leads to a better test result in comparison with the random method

and therefore increases the modeling efficiency.