Lift (data mining)

From Wikipedia, the free encyclopedia

In data mining, lift is a measure of the performance of a model at segmenting the population. The lift of a subset of the population is defined as the predicted response rate for that subset divided by the predicted response rate for the population.

For example, suppose a population has a predicted response rate of 5%, but a certain model has identified a segment with a predicted response rate of 20%. Then that segment would have a lift of 4.0 (20%/5%).

Traditionally, the modeller would seek to divide the population into ten deciles, and rank the deciles by lift. The marketer can then consider each decile, and by weighing the predicted response rate (and associated profit) against the cost, they can decide whether to market to that decile or not.