关键词: 关键词: 分类器, 改进Adaboost, BP神经网络, 组合模型, 权值调整, 归一化因子

Abstract: To solve the problem that it has poor generalization ability and low precision in the existing classification algorithm. the multi- classification combination algorithm was proposed, The model had integrated the neural network weak classifiers, the weak classifiers were distrbuted the weightand those classifiers were combined linearly into the strong classifier.In the process of weight iteration, the target of improved Adaboost algorithm was optimization the normalization parameter, and adjusted the strategy of samples updating weight, so that, It could ensure decreasing the error bound estimation as the weak classifiers were increasing ,the strong classifier integrated had been used to improve the prediction accuracy of model. Seven levels predication model had established, it had selected the daily meteorological datas which was six stations of Jiangsu Province, the 12 relative factors had been selected for important factors about rainfall, Through several experiments, the results show that the improved Adaboost-BP model has better performance, it is suitable for station of 58259 especially, the accuracy of overall samples is 81%,in the seven rainfall class, the accuracy of the class-0 rainfall is the best, the accuracy of other classes prediction have been improved, theoretical derivation and experimental results show that the improved algorithm can improved the accuracy of prediction.

Key words: Keywords: classifier, improved Adaboost, BP neural network, combined model, weight adjust, normalization parameter