珠江口外陆架沉积物地球化学数据的稳健主分量分析

ROBUST PRINCIPAL COMPONENTS ANALYSIS OF GEOCHEMICAL DATA FOR SEDIMENTS FROM THE CONTINENTAL SHELF OFF PEARL RIVER MOUTH

少量异点的存在有可能严重歪曲主分量分析结果。而稳健主分量分析方法以马氏距离为基础,通过迭代逐渐准确地鉴别异点,给异点赋予较小的权或零权,从而削弱了异点的影响,给出较能代表总体真实特征的主分量。文中以珠江口外陆架现代海洋沉积物地球化学数据研究的实例表明,即使对于容量大于200的大样本,也有必要采用稳健方法以保证统计分析结果的可靠性。

The existence of a few outliers may cause severe biases in the principal components analysis. Methods of robust principal components analysis may reduce the influence of outliers by iteratively identifying and downweighting outliers on basis of their Mahalanobis' distances. This results in the principal components mole representative of the nature of the population. This paper briefly introduces, three commonly used methods of robust principal components analysis and demonstrates the effectiveness of these methods. Two case studies, one of uranium geochemical data and the other of modern marine sediment geochemistry,show that even for large sample with a size greater than 200,it is necessary to adopt robust statistic methods in order to ensure the quality of the analytical results.