2.1 General Structure

The input data with \(n = 95'412\) rows and \(p= 479\) columns is structured as follows: \(\mathbf{D} = \{\{\mathbf{x}_i,\mathbf{y}_i\}\}, i = 1 ... n, \mathbf{x} \in \mathbb{R}^{p-3}, \mathbf{y} \in \mathbb{R}^2\).

There are \(m = p-3 = 476\) explanatory features, two targets and one unique identifier for each example.

The \(m\) features are grouped into four blocks of information:

• Member database with personal particulars, interests and organization-internal information on examples: 81 features
• Characteristics of example’s neighborhood from the US census 1990: 286 features
• Promotion history: 54 features
  • Summary of promotions sent to an example in the 12 months prior to the current promotion
  • Sending dates and RFA status of promotions 13-36 months prior to current promotion
• Giving history: 57 features
  • Summary statistics
  • Responses to promotions 13-36 months prior to current promotion