Multivariate time series models based on Reports of Substandard Acts and Conditions for forecasting occupational risk in open-pit mining

Forecasting occupational risk in open-pit mining remains a challenge due to high operational variability and the limited integration of preventive reporting systems into predictive frameworks. The objective of this study was to develop and compare multivariate time series models based on Reports of Substandard Acts and Conditions to forecast daily occupational risk in open-pit mining operations. To this end, records of substandard acts and conditions collected between January 2025 and March 2026 were processed, resulting in a daily time series of 455 observations and the construction of a Daily Occupational Risk Index. The results revealed high daily variability in Reports of Substandard Acts and Conditions, ranging from 12 to 121 records, with a predominance of medium risk levels. Among the evaluated models, Autoregressive Integrated Moving Average with Exogenous Variables achieved the best predictive performance, with MAE = 22.90, RMSE = 29.56, MAPE = 36.47% and MSE = 873.98, outperforming Vector Autoregressive, Vector Autoregressive Moving Average with Exogenous Variables and Long Short-Term Memory models. The three-month forecast indicated a slightly increasing trend in the Daily Occupational Risk Index, from 75.1 in April to 80.9 in June 2026, remaining within the medium risk level. It is concluded that Reports of Substandard Acts and Conditions, when integrated into a multivariate temporal framework, represent a valuable data source for anticipating occupational risk trends and supporting preventive decision-making in open-pit mining operations.