IPUMS.org Home Page

BIBLIOGRAPHY

Publications, working papers, and other research using data resources from IPUMS.

Full Citation

Title: Predicting Unemployment Status: Submission for the 2022 MEBDI Machine Learning Competition

Citation Type: Miscellaneous

Publication Year: 2022

Abstract: I use a set of machine learning tools to predict next-year unemployment status for individuals in the Current Population Survey, using a subset of variables specified by the MEBDI competition judges. I use a Decision Tree algorithm, and two regularized linear regressions, and I improve the classification performance by combining these three classifiers using a hard-voting ensemble. This combined classifier is able to predict future unemployment in a holdout testing set with 70.95% accuracy and future non-unemployment (employment or not-in-labor-force) with 79.92% accuracy, for an average accuracy across classes of 75.44%. The most important predictive features are a person's current work/employment status. Essentially , the best signal that someone will be unemployed next year is that they don't have a job this year.

Url: https://static1.squarespace.com/static/62014b6e32679565e6c69891/t/62b75804fb85de505e403a97/1656182788777/WinslowMebdi22Summary.pdf

User Submitted?: No

Authors: Winslow, Robert

Publisher:

Data Collections: IPUMS CPS

Topics: Labor Force and Occupational Structure

Countries:

IPUMS NHGIS NAPP IHIS ATUS Terrapop