Average importance of a predictor variable

Variables that improve a model have a non-zero relative variable importance in an individual model. Variables that don't improve an individual model have a relative variable importance of 0. For a linear regression model or a binary logistic regression model, all variables in the individual model have a relative variable importance of 1.

For example, suppose that a multiple response optimization considers models for 4 response variables. An individual predictor has the following relative importances in 3 of the models: 75, 56, 44. The predictor is not in the fourth model. Then the average importance = (75 + 56 + 44 + 0)/4 = 43.75