Predicting demographics of anonymous readers

Challenge

Ekstra Bladet wanted to predict age group and gender for anonymous (non-paying) visitors based on what is known about registered users — with predictions kept continuously up to date for business development.

Solution

Classification models trained on the registered user base, automated end to end.

• Data preparation. Statistics computed on gender- and age-segmented user groups; the entire preparation process automated.
• Model training. Support-vector-machine models for gender (binary) and age group (multi-class), cross-validated and tested on independent data.
• Model exposure. Models exposed as stored objects, callable for prediction from an R docker container with simple function calls.

Results

• Models deployed in multiple production contexts
• Automated retraining on latest data
• Enables targeted marketing & content