Fig. 3

Evaluation of a random forecast model for predicting the best day of time of phone call A Ranking of the importance of variables in the random forest model. Age and time of day of phone call are individual-level variables and others are ZIP code-level variables. Importance is computed as the increase of RMSE when the focal variable is randomly permuted in prediction. The distributions were obtained from 20 independent permutations. B The predicted number of phone calls for each time interval using the random forest model. Outcomes (complete or not complete) were obtained from the real-world records in the validation data. C The change of completion rate using the best phone call time predicted by the random forest model in each ZIP code area. Positive values represent improved completion rate