基于Bootstrap方法上的新修正EM算法

对于成分数据的处理往往是进行对数比变换，这时，如果数据中含有零点或是近似零点，变换后就会出现负无穷的数据．因此，对于成分数据中零点或是近似零点的处理就成为成分数据分析中的一个重要的课题，文献中采用了基本线性回归的修正EM算法，文章将此方法进行了改进，提出了基于Bootstrap方法上的修正EM算法，这两种方法均是在迭代的初值给定步骤进行了改进．模拟研究对上述提到的两种方法进行了比较，结果显示此方法相比原有的修正EM算法，算法更为简便，得到的估计更为精确．

A New Modified EM Algorithm Based on the Bootstrap Method

For composition data processing is often for logarithmic than transform, at this time,if the data contains zero or approximate zero, transform and then be negative endless data. So ,for composition data in the processing of the approximate zero zero or become composition analysis of data is an important topic,literature used basic linear regression fixed EM algorithm, this paper the method is improved, is based on the Bootstrap method. The methods are the initial value given in the iteration step is modified. Simulation research and empirical research on the above mentioned three methods are compared, the results showed that the method in the paper compared the original modified EM algorithm method is simple,and get more accurate estimate.