Machine Learning Risk Assessments in Criminal Justice Settings by Richard Berk

Author:Richard Berk
Language: eng
Format: epub, pdf
ISBN: 9783030022723
Publisher: Springer International Publishing

5.5.2 Variable Importance for Random Forests

Especially when used for forecasting, the most important output from classifiers is the classes assigned to different observations. But for criminal justice decision makers, the forecasts alone may not be sufficient. Because of legal and administrative concerns, knowing the importance for classification of each predictor can be very instructive. How important, for instance, is gender? What about age or prior record?

A useful operationalization of a predictor’s importance is its contribution to classification accuracy. For random forecasts, this is handled in a clever manner. Classification accuracy is first computed using all of the available predictors. Then, each predictor is altered one at a time so that it cannot contribute to classification accuracy. The resulting drop in classification accuracy for each predictor is a measure of importance. Here is the algorithm classification importance.6 1.Grow a tree in a random forest as usual.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.