Predictive Analytics: Data Mining, Machine Learning and Data Science for Practitioners by Dursun Delen

Author:Dursun Delen [Delen, Dursun]
Language: eng
Format: epub
Tags: Business & Economics, statistics, Production & Operations Management, computers, Data Science, Data Analytics
ISBN: 9780136738510
Google: v7ZrzQEACAAJ
Publisher: Pearson Education
Published: 2020-10-30T00:31:39.148550+00:00

Transparency

The interpretation of ensembles can become quite difficult. If you build a random forest ensemble containing 200 trees, how do you describe why a prediction has a particular value? You can examine each of the trees individually, though doing so is clearly not practical. For this reason, ensembles are often considered black-box models, meaning that what they do is not transparent to the modeler or domain expert. However, you can look at the split statistics (which variables are more often picked to split early in those 200 trees) to artificially judge the level of contribution (a pseudo variable’s importance) each variable makes to the trained model ensemble. Compared to a single decision tree, this is still difficult, and it is not a very intuitive way to interpret how a model performs. Another way to determine which inputs to the model are most important is to perform a sensitivity analysis.

In addition to the complexity and transparency issues, model ensembles can also be difficult and computationally expensive to build and deploy. Table 6.1 shows the pros and cons of model ensembles as they compared to individual models.

Table 6.1 Advantages and Shortcomings of Model Ensembles

Advantages

Description

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.