Performance Assessment of Regression and Neural Network Models for Structural Steel Weight Prediction in Industrial Buildings

The strength, durability, and speed of steel construction make it a leading material for industrial buildings. Reliable cost and schedule estimates are supported by accurate and early predictions of steel weight per unit area (tons/m²). This study used 180 historical projects constructed in Indonesia (2010–2024) to compare the use of artificial neural networks (ANNs) and multiple regression analysis (MRA) in predicting steel weight per unit area in industrial buildings. The collected data were the actual construction of 180 projects of industrial buildings with various parameters that influence the weight of steel structures, such as span, length, bay spacing, height of eaves, rafter pitch, structure type, and type of roof. The results showed that in terms of mean squared error (MSE), mean absolute percent error (MAPE), R2 value, and prediction accuracy (PA), the ANN model significantly outperformed the regression model.