Rapid and dramatic fixes of a broken propensity model:

A legacy customer acquisition propensity model needs improvement and poor R2 scores are increased more than ten-fold.

Key changes fuelling the improved model performance included; improving data pipelines and the number of training instances by 54%, aligning the choice of predictor variables to best practice for this sector, abandoning the linear logistic model for better fit decision tree options.

A more detailed project synopsis can be downloaded: from the following link:

content_acqPred_tenXImprove.pdf

Download PDF • 739KB