Regression Technique for Pure Component Data

To perform a regression means to adjust a correlation function to given data points in a way that the representation of the data points is as good as possible. This is achieved by minimizing an objective function F. Consider m to be a general pure component property, this objective function is usually defined as 

The summation of the squares of the deviations has the target to eliminate the influence of the sign of the deviations and to emphasize the points with large deviations. The weighting factor can sort out obviously poor data points or assign a special weight to individual data points, for example, according to their experimental imcertainties. Many other forms for the objective function are possible, for example, the sum of relative or absolute deviations. 

1) The chosen correlation might not be capable to reproduce the data satisfactorily with the required accuracy. In this case, a better one, maybe with a larger number of adjustable parameters, has to be chosen. Currently, there are correlations available for each property which are finally capable enough to reproduce any reasonable data far a specified component in the temperature range of interest. 

2) Some data points or whole data sets can be wrong or not accurate enough. In this case, they should be removed from the database, for example, by setting their weighting factor to 0. To decide which data points have to be removed, the following procedure has proved to be successful  

1) The following considerations are general ones, independent from the objective function. 

---

I Appendix E Regression Technique for Pure Component Data... 

---