中文摘要：

研究了多类别样本数据集的分类，针对传统的“一对一”或“一对多”BP．Ada．Boost算法，训练时间开销随着训练样本数以及训练样本种类的增加急剧增加，使其实际应用十分受限，尤其不适用于大规模数据分类的问题，提出了将多分类BP神经网络与使用多类分类指数损失函数的逐步叠加建模（SAMME）算法相结合以构造AdaBoost强发类的Multi．BPAdaBoost算法，实现模型信息的有效利用与融合增强。对传统“一对多”BP．AdaBoost算法和Multi．BPAdaBoost算法进行了对比试验，结果表明，在相同测试情况下，后者有效降低了BP．AdaBoost训练过程中的时间开销。

英文摘要：

The study focused on the classification of the dataset referred to multi-class samples, and paid attention to the problem that the time cost of traditional ＂one-against-one＂ or ＂one-against-all＂. BP-AdaBoost algorithm increases rapidly with the increase of the sample amount and the sample class number, thus leading to the hindrance to its practical application, especially when dealing with large-scale datasets. Then, to solve this problem, the muhi-BP- AdaBoost algorithm was proposed by combinig multi-class BP neural networks with the algorithm of Stagewise Additire Modeling using a Multi-class Exponential loss function （SAMME） to construct a strong AdaBoost classifier. The algorithm can effectively use and fuse model information to improve its performance. The test on the traditional ＂one-against-all＂ BP-AdaBoost algorithm and the proposed muhi-BP-AdaBoost algorithm was performed, and the results showed that the latter had the better abiligy in reducing the time cost than the former under the same testing conditions.