Abstract: The neural network model is used to learn and simulate the ground state spin distribution of the nucleus under stochastic two-system ensemble (TBRE), and the input characteristics of the learned model are analyzed. This is a typical application of classification using neural network models in nuclear physics. We show that it is still difficult to accurately each the sample within random interaction ensemble using the single hidden layer neural network model in this paper. However, the NN model describes the statistical properties of the ground state spins reasonably well, probably because the NN model learned the empirical law of the ground state spin distribution in TBRE.