Forecasting occupational accidents using artificial intelligence-based machine learning methods: A 2025–2033 projection based on Türkiye's 2016–2024 SSI data

The primary objective of this study is to predict trends in temporary incapacity and fatal occupational accidents for the 2025–2033 period, based on SGK occupational accident data from 2016–2024, to generate proactive future projections and compare the performances of various algorithms. Within the scope of the methodology, the dataset was organized based on injury type codes and structured as a multi-target framework including temporary incapacity variables for 1, 2, 3, 4, and 5 or more days, as well as variables for deceased male and female workers. Following the data preprocessing and the generation of lag features, Random Forest, XGBoost, and LightGBM algorithms were tested, and model performances were evaluated using MAE and RMSE metrics. According to the findings, the Random Forest model produced the lowest error values (MAE: 387.81, RMSE: 933.77) among the compared models. Based on this primary model, the annual total number of occupational accidents, expected to be 400,604 in 2025, is calculated to reach 639,786 in 2033, representing a 59.7% increase. While "5 or more days" reported cases constitute the highest volume (increasing from 248,007 in 2025 to 395,207 in 2033), it was determined that a total of 63,432 employees will lose their lives during the 2025–2033 period, with this mortality projection concentrated largely among male workers with 61,871 cases. In conclusion, it has been proven that the Random Forest algorithm provides a robust decision support tool for data-driven risk analysis, and it was established that injuries with a duration of five days or more and male worker mortality should be priority areas for intervention in occupational health and safety policies.