Experimentation with cost-sensitive boosting of the absolute loss [1] in the component-wise boosting framework [2]. This is particularly useful if over- and underestimation of the outcome should be penalised differently.

References

[1] B. Kriegler and R. Berk, ‘Small area estimation of the homeless in Los Angeles: An application of cost-sensitive stochastic gradient boosting’, Ann. Appl. Stat., vol. 4, no. 3, Sep. 2010, doi: 10.1214/10-AOAS328.

[2] P. Bühlmann and T. Hothorn, ‘Boosting Algorithms: Regularization, Prediction and Model Fitting’, Statist. Sci., vol. 22, no. 4, Nov. 2007, doi: 10.1214/07-STS242.