Integrated OVMD-BiGRU-SMAC Framework for Forecasting Construction Accidents in the Kingdom of Saudi Arabia

Construction accidents remain a major concern for occupational safety, especially within high-risk environments such as those found across the expanding construction sector in the Kingdom of Saudi Arabia (KSA). This study presents a novel forecasting framework that integrates signal decomposition strategy with deep learning to model and predict construction accident occurrences. The approach begins with the application of Optimized Variational Mode Decomposition (OVMD) to extract meaningful temporal components, referred to as Intrinsic Mode Functions (IMFs) from the historical construction accident time series data. Each IMF is then modeled independently using a Bidirectional Gated Recurrent Unit (BiGRU) network, which captures complex temporal dependencies in both forward and backward directions. In order to obtain optimal forecasting performance, Sequential Model-based Algorithm Configuration (SMAC) is employed to fine-tune the hyperparameters of each BiGRU model. The proposed framework is trained on monthly construction accident data in the KSA from June 2010 to March 2024. Among the tested configurations, the proposed OVMD–BiGRU–SMAC model produced the most reliable and better results and achieves RMSE value of 17.26, MAE of 14.02, and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.874. In comparison, the OVMD–TCN–SMAC model showed the weakest performance, with an RMSE of 20.27, MAE of 16.72, and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.742. These results demonstrate the effectiveness of combining signal decomposition with deep learning techniques in order to caputer the irregular and nonstationary patterns of construction accident data and provide more reliable forecasts to support safety management and proactive planning efforts.