Accuracy in predicting assay properties can be measured by two common statistics: absolute prediction error, and the R^2 value. The R^2 factor is the fraction of the original variability of the raw data that is removed by the correlation. A value of 0.0 says that the prediction error is the same as the variability of the raw data (i.e. no improvement), while a value of 1.0 says that the prediction exactly matches the measured results (i.e. very good).

The prediction errors are calculated for up to 6 cuts to separate the effects of cut points from the effects of Whole Crude properties. For simplicity, the average errors for the "World" model over the actual cuts are shown here.

The "World" model includes crudes from all over the world. Other models have been made, such as for Saudi Arab crudes, and they show better accuracy. These models, based on subsets of assays, can be made for special uses.

The accuracy predictions are delivered with the product in several forms. The documentation for the CCPP models shows the accuracy both graphically (below) and in tabular form (values in 6 Avg Boiling Pt temperature ranges). The CCPP models have the same table internally, and it is used to provide model accuracy when requested.

The red triangles are the results for a particular crude. The tuning process makes the model exactly match the Base assay. The tuning factors are small adjustments to K, or property offsets.