Unlocking telecommuting patterns before, during, and after the COVID-19 pandemic: An explainable AI-driven study

Abstract

The COVID-19 pandemic has instigated a global paradigm shift in employment practices, precipitating a widespread transition to telework. While past events had no long-lasting effect on the continued working conditions of the population, it is unclear what a prolonged need for telecommuting on such a nationwide scale would continue to have on the working population. This study uses an explainable artificial intelligence approach to investigate the changes in those telecommuting across three periods: i) pre-pandemic, ii) pandemic, and iii) post-pandemic periods. Machine learning methods, including decision trees, random forest, extreme gradient boost, naïve Bayes, and artificial neural networks, were developed across the study periods. Shapely Additive Explanations, a model-agnostic approach, explains the best-performing model for each period. Results show that before the pandemic, gender and occupation were major determining factors of telecommuting adoption. However, the pandemic reduced the impact of these factors, making income and education levels a more significant factor for identifying telecommuters. Additionally, the study examines interaction effects between these features, allowing for a deeper investigation of specific aspects of interest. These insights can be instrumental in shaping policies surrounding telecommuting as the pandemic gradually subsides. By understanding the changing dynamics of telework, decision-makers can better support and adapt to the evolving needs of the working population.

Keywords

• Explainable artificial intelligence
• Telecommuting
• Pandemic
• Machine learning
• Shapely additive explanations