针对有噪声的高维数据引起决策树预测准确率下降的问题,利用容噪主成分分析(Noise-free Principal Component Anlysis,NFPCA)算法思想对C4.5算法改进而形成NFPCA-in-C4.5算法。该算法一方面将高维数据噪声控制问题转化为拟合数据特征与...针对有噪声的高维数据引起决策树预测准确率下降的问题,利用容噪主成分分析(Noise-free Principal Component Anlysis,NFPCA)算法思想对C4.5算法改进而形成NFPCA-in-C4.5算法。该算法一方面将高维数据噪声控制问题转化为拟合数据特征与控制平滑度相结合的最优化问题,从而获得主成分空间;另一方面在决策树自顶向下构建新节点的过程中,再将主成分空间恢复到原始数据空间来避免降维过程中属性特征信息永久消失。实验结果表明NFPCA-in-C4.5算法兼具降维和容噪功能,避免了降维中由特征信息损失和噪声残留造成的预测模型准确率大幅降低的问题。展开更多
Based on the tropical cyclone(TC) observations in the western North Pacific from 2000 to 2008, this paper adopts the particle swarm optimization(PSO) algorithm of evolutionary computation to optimize one comprehensive...Based on the tropical cyclone(TC) observations in the western North Pacific from 2000 to 2008, this paper adopts the particle swarm optimization(PSO) algorithm of evolutionary computation to optimize one comprehensive classification rule, and apply the optimized classification rule to the forecasting of TC intensity change. In the process of the optimization, the strategy of hierarchical pruning has been adopted in the PSO algorithm to narrow the search area,and thus to enhance the local search ability, i.e. hierarchical PSO algorithm. The TC intensity classification rule involves core attributes including 12-HMWS, MPI, and Rainrate which play vital roles in TC intensity change. The testing accuracy using the new mined rule by hierarchical PSO algorithm reaches 89.6%. The current study shows that the novel classification method for TC intensity change analysis based on hierarchic PSO algorithm is not only easy to explain the source of rule core attributes, but also has great potential to improve the forecasting of TC intensity change.展开更多
文摘针对有噪声的高维数据引起决策树预测准确率下降的问题,利用容噪主成分分析(Noise-free Principal Component Anlysis,NFPCA)算法思想对C4.5算法改进而形成NFPCA-in-C4.5算法。该算法一方面将高维数据噪声控制问题转化为拟合数据特征与控制平滑度相结合的最优化问题,从而获得主成分空间;另一方面在决策树自顶向下构建新节点的过程中,再将主成分空间恢复到原始数据空间来避免降维过程中属性特征信息永久消失。实验结果表明NFPCA-in-C4.5算法兼具降维和容噪功能,避免了降维中由特征信息损失和噪声残留造成的预测模型准确率大幅降低的问题。
基金National Natural Science Foundation of China(41201045)Jiangsu Qing Lan Project(2016)Natural Science Foundation of Jiangsu Province(BK20151458)
文摘Based on the tropical cyclone(TC) observations in the western North Pacific from 2000 to 2008, this paper adopts the particle swarm optimization(PSO) algorithm of evolutionary computation to optimize one comprehensive classification rule, and apply the optimized classification rule to the forecasting of TC intensity change. In the process of the optimization, the strategy of hierarchical pruning has been adopted in the PSO algorithm to narrow the search area,and thus to enhance the local search ability, i.e. hierarchical PSO algorithm. The TC intensity classification rule involves core attributes including 12-HMWS, MPI, and Rainrate which play vital roles in TC intensity change. The testing accuracy using the new mined rule by hierarchical PSO algorithm reaches 89.6%. The current study shows that the novel classification method for TC intensity change analysis based on hierarchic PSO algorithm is not only easy to explain the source of rule core attributes, but also has great potential to improve the forecasting of TC intensity change.
