Training

When building models with earlier versions of Clementine, there was no way for you to know in advance which technique would produce the most accurate model. You had to test each model manually and then compare results.

Now, with Clementine® 11.0, you can automatically pick the best model if you are predicting a binary (yes/no) outcome. The Binary Classifier node enables you to create and compare models for binary outcomes using a number of different methods, making it easier to try out a variety of approaches and quickly compare the results. Select the modeling algorithms that you would like to use and the specific options for each. You can also specify multiple variants for each model. For example, when creating Neural Net models, you no longer have to choose between the quick, dynamic, or prune methods—ou can try them all at once

Take the following steps to use the Binary Classifier node:

• Add a Binary Classifier node to the stream and connect if to the Type node. You can also attach it to your Source node if that is where you type your fields.
• Edit the Binary Classifier node and click the Fields tab
• Click the Use custom setting button
• Select your Target field (in this case, "Loyal") and all the fields you want to be considered as potential predictors

The Model tab allows you to specify the number of models to be saved, along with the criteria used to compare models. With the Expert tab, you can choose which modeling types you want to try and modify the model parameters.

In the example above, all the modeling techniques have been requested along with the default model parameters. Activate the Binary Classifier node by clicking on the Execute button.

The output in the example above has been sorted according to overall accuracy.

We see that the C 5.0 model had the best overall accuracy of 88.267%. If this is the desired model, check the box to the left of the C 5.0 model and use the Generate menu at the top of the output to generate the actual model, an evaluation chart, or a copy of the modeling node with all the parameters that were used to build the model.