【摘要】 结合TF-IDF算法思想,提出了特征频率、森林频率以及伪梯度提升决策树,解决了梯度提升决策树随着迭代次数的增加,错误数据被边缘化的问题。在伪梯度提升决策树中,所有决策树分别在原始数据集的Bootstrapping后的数据集上产生,无须针对每次迭代来对数据集采样。在分布式集群上进行内网防御的实验,结果表明在一定规模的训练集上,伪梯度提升决策树具有更好的预测准确度。更多还原

【Abstract】 Combining with the idea of TF-IDF algorithm,the frequency of characteristics(Eigen Frequency),the frequency of forest(Forest Frequency)and the pseudo boosting decision tree(PBDT)were put forward,solving the marginalized problem of wrong data with the increasing number of iterations for gradient boosting decision tree(GBDT).In PBDT,all the decision trees produce respectively in data sets after the original data set of the Bootstrapping,without aiming at each iteration to sample data sets.Then intranet defense experiment was conducted on distributed cluster.The experimental results show that on the training set with a certain scale,PBDT has better prediction accuracy.更多还原