Predicting declining and growing occupations using supervised machine learning

In the United States (U.S.), structural changes in the economy remain varied, yet continuous, prompting the need for regular analyses of both declining and growing occupations. As automation, robotization, and digitization continues to accelerate and drive new patterns of economic change, so does the need for proactive programs and policies aimed at targeted workforce re-training. Applying machine learning (ML) to occupational data provides one potential approach to inform such workforce initiatives, specifically by helping to predict both declining and growing occupations with advanced accuracy. In this paper, we examine the extent to which occupational attributes are predictive of the declining and growing status of jobs in the State of Ohio (USA). In particular, we examine the results from five distinct supervised ML models (i.e., multinomial logistic regression, nearest neighbors, random forest, adaptive boosting, and gradient boosting), and data on the characteristics of occupations from O*NET, as well as information on employment changes from the U.S. Bureau of Labor Statistics. We found that the random forest and gradient boosting models perform the best, predicting declining and growing jobs in Ohio at roughly 92% accuracy in the test set. Moreover, our analysis revealed that the most important features in predicting declining occupations are physical (e.g., spending time making repetitive motions), while the most important features in predicting growing occupations are related to obtaining information and communication. Our method can be replicated at a local or regional level to help practitioners predict future occupational shifts, ultimately enhancing economic and workforce development efforts.