Research Article
Machine Learning-Based Forecasting of U.S. State Unemployment Rates for 2026
@INPROCEEDINGS{10.4108/eai.22-5-2026.2365158, author={Jiahao Ma and Tingyu Xie}, title={Machine Learning-Based Forecasting of U.S. State Unemployment Rates for 2026}, proceedings={Proceedings of the 4th International Conference on Image, Algorithms, and Artificial Intelligence, ICIAAI 2026, 22-24 May 2026, Singapore, Singapore}, publisher={EAI}, proceedings_a={ICIAAI}, year={2026}, month={8}, keywords={Unemployment; State-level forecasting; Machine learning; Panel data; GDP growth}, doi={10.4108/eai.22-5-2026.2365158} }- Jiahao Ma
Tingyu Xie
Year: 2026
Machine Learning-Based Forecasting of U.S. State Unemployment Rates for 2026
ICIAAI
EAI
DOI: 10.4108/eai.22-5-2026.2365158
Abstract
This paper forecasts U.S. state unemployment rates for 2026 using quarterly data from 2020–2025 and a macroeconomic predictor, real GDP growth by state. Descriptive statistics document the 2020 pandemic shock, the recovery through 2022, and a mild uptick during 2023–2025, alongside large cross-state differences in peak unemployment. To quantify the output–labor link, a fixed-effects regression for Michigan and Hawaii indicates a countercyclical relationship between GDP growth and unemployment (β≈−0.48; R²≈0.20). For prediction, linear regression is compared with random forest and gradient boosting. In cross-validation and a 2025 hold-out test, the linear model achieves the lowest mean squared error, while the ensemble models show signs of overfitting. Using a ±0.25 percentage-point rule relative to end-2025, most 2026 forecasts are classified as stable.
