Data Mining - Professional SQL Server Analysis Services 2005 with MDX - page 748

Figure 14-15
That's all well and good, but how is the lift chart calculated and visually rep-
resented? Analysis Services 2005 predicts the bike buyer attribute for every
row in the input table. Each prediction also has a value called predict probab-
ility that provides a confidence value associated with the prediction. If you
have a predict probability of 1.0, that means that the model believes the pre-
dicted value is always correct. If the predict probability for an input is 0.90,
that means there is a 90% probability the predicted value is correct. After pre-
dicting the values for all the input rows, the results are ordered based on the
predict probability. Now the predicted results are compared with the original
value to calculate the prediction correctness percentage, and then they are
plotted on a graph. The input data set is ordered based on the predict probab-
ility for the predicted value. This can be seen in the graph in Figure 14-15
where the lines indicating the current model and ideal model are nearly
identical up to 10% of the population. The mining legend shows the score,
population percentage, and predicts probability values for a specific popula-
tion selection. Figure 14-15 shows that the population selection is 50%, which
is indicated by a darker line. You can select a different population percentage
by clicking the mouse on a specific population. When 100% of the data set is
considered, the decision tree mining model is able to predict correct values