Adapting Your Models

It is important to assess the quality of a model and adapt it to optimize results.

You can assess the ability of a model to be accurate at point of creation. It needs to be of high quality and robust, so it can be re-applied to future data sets. Two ways you can assess models are:

1. The number of records in the sample set.
2. The number of Predictor variables, otherwise known as Table columns, that are used to build the model.

The Predict Inclusion tool is aimed at Marketers and therefore the permutations for analysis of model parameters are limited, but some defaults have been set in the configuration that should be used as guidelines in relation to the grid above.

Sample Size

The default is 5000, this is the number of records from the target group that is used to build the model. The model process will take a maximum of 2,500 rows that represent the target (AND universe) domain and 2,500 non-target (AND universe) domain. Campaigns can handle less that 5000 records, but care must be taken not to have too small a target, as this rapidly reduces model solidity.

Max Model Columns

The default of 75 is the number of columns to be used as predictor columns in building the model. If you build a model with only 5 - 10 columns or predictor variables, it is likely that some variables with stronger explanatory power could be missed.

It is important to note that if, for example, you use the 'Age' predictor variable column, you would not add any additional information to the model if you included 'Age Band'. The tool will take the Age column and, as it does with all columns, analyze it for most appropriate treatment. Age would probably be classified as a continuous predictor variable. A continuous predictor variable is a numeric column with a high cardinality of values that exist in its range. As it has a continuous range of values, it is analyzed for naturally existing bands. The marketing convention bands such as 'Under 25', or '25-35' are unlikely to add values to the model as they are based on the same underlying data in the Age column.

In this way the number of columns added to the model will be augmented by further data engineering performed by the model building process when contributing to the 75 max columns to be used.

You can also add powerful data engineering nodes to the model, such as aggregates and expressions, you are not restricted to objects that appear as database columns.

• Modeling Functionality Overview
  • Comparison Between Modeling Tools
  • Adapting your Models