Test of the scoring model. Analysis of the final test and its values

After completing the creation of a test of scoring model, it is necessary to analyze it and determine whether the model is good enough or whether it still needs to be refined and rebuilt, and also to determine what decisions can be made based on this model.
Learn more about the testing process here.

First of all, you need to pay attention to the final Gini Index. It shows the predictive and gradual strength of the model. Namely, how strong is the dependence of the model in relation to the fact that the more points the client gains, the higher the probability that he will be good.
If the Gini index is lower than 30%, then such a model is considered to be of poor quality at all. And it makes sense to try to rebuild it again.
If the Gini index is above 30%–40%, then such a model is already considered worthy and can be used.
If your Gini index is more than 60%, then you can safely be sure that the model is very, very strong.
The next step is to determine the threshold for accumulating points, up to which the system will refuse in one or another process for the client.It depends more on what goals were pursued to create a scoring model, as a rule, it is necessary to determine the threshold of bad clients, which the user is ready to skip based on his requirements, and then take the minimum number of scores from this line as the minimum threshold for passing the scoring model.
With the columns with the "Accumulated ..." prefix, you can see how many and which clients will be rejected in total in quantitative or better in percentage terms at the selected minimum threshold.
Model analysis is easier making for most users in excel. To do this, you can export both the scoring model and the test to an excel file.More details about the data exported to excel can be found here.
As soon as a scoring model is created that has come up in terms of its predictive strength, the minimum threshold for passing is selected, you can implement the scoring model in your system.
Calibration of any scoring model is still needed periodically. Usually every 3 months, because products, marketing strategies, customer segments, macroeconomic features and more can change. Many factors can change the behavior of the client and that is why do not forget to check the scoring model for relevance through the same testing and calibrate or completely change it and create a new model.
Scoring Machine will help you with this.
If you have any questions about using the system, feel free to contact support.