相幅组合的函数型数据特征提取方法研究

Feature extraction for functional data by combining phase and amplitude information

当前函数型数据分析更多关注于函数的振幅变化而忽略相位变化,很多场合下,相位变化中含有对统计分析有用的信息。基于偏最小二乘法提出了相幅组合的函数型数据特征提取方法,首先使用函数对齐技术获得刻画相位变化的时间弯曲函数,再将对齐函数和弯曲函数通过分段函数的方式重新组合,最后利用偏最小二乘法提取相幅组合函数的成分特征,并应用在回归和分类模型上。实验结果表明,与主成分分析方法相比,所提方法具有更优越的预测性能。

The current functional data analysis focuses on the amplitude variability,while ignoring the phase variability.In many cases,phase variability is a kind of useful information for statistical analysis.Based on partial least square method,this paper proposed a feature extraction procedure for functional data by combining phase and amplitude information.Firstly,the proposed procedure used a function alignment technique to obtain the time warping function containing phase variation.Secondly,it combined the aligned function and the warping function in the form of piecewise function.Finally,it used the partial least square approach to extract component features from the combined function and utilized those features on the regression and classification model.Experimental results show that the proposed method can obtain better prediction performance than the principal component analysis method.