神经网络算法是一种非常经典的分类算法,然而神经网络的一个不足之处就是容易陷入过拟合。针对这种不足,正则化神经网路算法(RNM)与提前终止迭代算法被提了出来。为了进一步研究这两种算法性能的差异,本文在包括金属岩石分类数据集(sonar数据集)在内的20个UCI标准数据集上对这两种方法进行了性能测试。实验显示在分类准确率上正则化神经网路算法要更优秀一些,但是在分类速度上提前终止迭代算法更占优势。 Neural network algorithm is a very classic classification algorithm, but one of the disadvantages of neural network is easy to fall into over-fitting. In response to this problem, regularized neural network (RNM) and early termination iterative algorithm are proposed. In order to further investigate the difference between the performance of the two algorithms, the paper tests the performance of the two methods on 20 UCI standard datasets including metal rock classification data set (sonar data set). Experiments show that the algorithm of regularization neural network is better in classification accuracy, but it is more dominant to stop the iterative algorithm ahead of the classification speed.