Classification Trees with Mismeasured Responses

Liqun Diao () and Grace Y. Yi ()
Additional contact information
Liqun Diao: University of Waterloo
Grace Y. Yi: University of Western Ontario

Journal of Classification, 2023, vol. 40, issue 1, No 8, 168-191

Abstract: Abstract Classification trees are a popular machine learning tool for studying a variety of problems, including prediction, inference, risk factors identification, and risk groups classification. Classification trees are basically developed under the assumption that the response and covariate variables are accurately measured. This condition, however, is often violated in practice. Ignoring this feature commonly yields invalid analysis results. In this paper, we study the impact of mismeasured responses on the performance of standard classification trees and propose a novel classification trees algorithm for mismeasured responses. Our study is directed to settings with binary responses which are subject to mismeasurement. To address the effects of mismeasured responses, we modify the decision rules which are valid for tree building in the mismeasurement-free settings by introducing new measures for the node impurity and misclassification cost. To characterize the magnitude of mismeasurement in responses, we consider two data scenarios. In the first scenario, the mismeasurement rates are known, either from previous studies of the same nature or being set by researchers who are interested in conducting sensitivity analyses to assess the impact of mismeasured responses. In the second scenario, the mismeasurement rates are unknown and are estimated from a validation dataset which contains both accurate measurements and mismeasurements for responses. We conduct a variety of simulation studies to assess the performance of the proposed classification trees algorithm, in comparison to the usual classification trees algorithms which ignore response mismeasurement. It is demonstrated that ignoring response mismeasurement can yield seriously erroneous results and that our proposed method provides superior performance with the mismeasurement effects accommodated. To illustrate the usage of the proposed method, we analyze the data arising from the National Health and Nutrition Examination Surveys (NHANES) by conducting sensitivity analyses to assess how classification results may be affected by different misclassification costs.

Keywords: CART; Classification trees; Mismeasurement in response; Mismeasurement probability; Sensitivity analyses; Validation data (search for similar items in EconPapers)
Date: 2023
References: View references in EconPapers View complete reference list from CitEc
Citations:

Downloads: (external link)
http://link.springer.com/10.1007/s00357-023-09430-6 Abstract (text/html)
Access to the full text of the articles in this series is restricted.

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:spr:jclass:v:40:y:2023:i:1:d:10.1007_s00357-023-09430-6

Ordering information: This journal article can be ordered from
http://www.springer. ... hods/journal/357/PS2

DOI: 10.1007/s00357-023-09430-6

Access Statistics for this article

Journal of Classification is currently edited by Douglas Steinley

More articles in Journal of Classification from Springer, The Classification Society
Bibliographic data for series maintained by Sonal Shukla () and Springer Nature Abstracting and Indexing ().