To do or not to do? Cost-sensitive causal classification with individual treatment effect estimates

• We introduce a cost-sensitive decision-making framework for causal classification. derive the classification boundary to maximize expected profit. establish new ranking approach with individual treatment effects. Experiments demonstrate an increase in profit compared cost-insensitive ranking. Individual effect models allow optimizing by predicting of on outcome interest instances. These predictions selecting instances treat order optimize overall efficiency and net effect. In this article, we extend upon value based effects case binary outcome. The is linear function estimated effect, positive probability cost benefit parameters problem setting. It allows causally classifying negative class profit, which introduced as objective at hand present ranker ranks maximizing each possible threshold that results from constraint number treatments differs conventional proposed experimentally evaluated synthetic marketing campaign data sets. indicate presented method outperforms approach.