Abstract:
The article considers the problem of forecasting the values of one variable from the values of another variable using regression analysis techniques. The list of compared regression analysis procedures included the following parameter estimation methods: ordinary least squares, least squares based on the sign statistics, and zero correlation based on rank statistics. To implement the method of least squares on the basis of sign statistics, it is necessary to construct efficient data processing procedures. An efficient variant of the corresponding algorithm was realized for a difficult situation, when the goal function is piecewise constant. The comparative analysis of regression analysis procedures in real conditions, when data are not normally distributed, showed that nonparametric techniques are advantageous. In this case, taking into account algorithmic aspects, the preference should be given to procedures based on rank, not on signs. Advantages of nonparametric methods can improve the accuracy of measuring chromogranin A, widely used as an immunohistochemical marker of neuroendocrine differentiation.
Keywords:regression analysis; rank and sign-based procedures; prediction quality; adjustment of measurement results.
Fulltext:
PDF file (212 kB)