Tehran University of Medical Sciences International Journal of Occupational Hygiene 2008-5109 17 3 2026 06 08 138 153 Elahe Jafari Department of Occupational Health Engineering, Faculty of Health, Tehran Medical Sciences, Islamic Azad University, Tehran, Iran Hamzeh Mohammadi Department of Occupational Health Engineering, Faculty of Health, Tehran Medical Sciences, Islamic Azad University, Tehran, Iran Majid Bayatian Department of Occupational Health Engineering, Faculty of Health, Tehran Medical Sciences, Islamic Azad University, Tehran, Iran Maryam Behboudi Department of Statistics, Science and Research Branch, Islamic Azad University, Tehran, Iran Davod Panahi Department of Occupational Health Engineering, School of Public Health and Safety, Shahid Beheshti University of Medical Sciences, Tehran, Iran 2024 12 07 2025 09 16 Background: Work-related injuries data analysis helps the choice of measures to prevent accidents, therefore, the information and data of occupational injuries were analyzed to identify the overall trend of occupational injuries, and develop a forecasting model of lost workdays and provinces clustering. Methods: To achieve the first goal, we calculated NFIR and FIR per 100000 workers and injury indices (including AFR, ASR, FSI, Safe T-Score, IR, MDR, and LTIR). To reach the second purpose, the FEE, FTE, FETE, and RE linear models and supervised machine learning alongside linear models (Random Forest, Extra trees, XG Boost, G Boost) were used. Finally, the AP clustering algorithm for provinces clustering, time series clustering (DTW method), and the KNN forecasting algorithm were applied. Data for 378826 occupational injuries, which occurred from 2001 to 2019, were extracted from the publications of the ISSO. Industries data were extracted from the Ministry of Industry publications. Results: NFIR to FIR ratio ranged between 60 to 265.35 and injuries indices increase from 2001 to 2008 and then experienced a decline. In linear models, OLS and RE had the best performance in forecasting the loss days and the extra trees method had better performance as a blender than random forest method. The clustering results showed that Khuzestan, Markazi, Mazandaran, Qazvin, and Tehran provinces are in cluster 1 and other provinces are in cluster 2. Conclusion: This study can be regarded to forecast occupational injuries and safety promotion planning. https://ijoh.tums.ac.ir/index.php/ijoh/article/view/660 https://ijoh.tums.ac.ir/index.php/ijoh/article/download/660/877