Search

The precise prediction of construction costs during the initial phase of a construction project is crucial for ensuring the project’s success. Identifying the parameters that influence project cost contributes to achieving accurate results and improves the overall accuracy of cost estimation. This study applied three machine learning methods such as artificial neural network (ANN), natural gradient boosting (NGBoost), and linear regression (LR) models, to predict the total cost of construction. The NGBoost model was employed for construction cost estimation and was compared with two machine learning algorithms: artificial neural network and linear regression. Evaluation metrics, including Mean Absolute Error (MAE), Coefficient of efficiency (CE), Mean Absolute Percentage Error (MAPE), index of agreement (d), and coefficient of determination (R2), are employed to assess and compare the accuracy of the developed algorithms. Statistical indicators revealed that the NGBoost algorithm outperformed others, displaying the highest coefficient of determination (R2= 0.992 for training and R2= 0.985 for testing) and the lowest root mean square error (RMSE = 0.5136 for training and RMSE = 0.3702). Moreover, sensitivity analysis revealed that the input parameter with the highest contribution was formwork, accounting for nearly 41%. On the other hand, the superimposed load had the lowest contribution, totaling 5%. The Shapley Additive Explanation (SHAP) method was employed to elucidate the importance and contribution of input variables influencing construction costs. The findings of this study offer valuable insights for project stakeholders, enabling them to minimize errors in estimated costs and make informed decisions early in the construction process.

Concrete compressive strength (CCS) is the most crucial structural engineering designing conventional concrete and high-performance concrete (HPC) structures. Accurately predicting high-performance concrete (HPC) compressive strength is crucial when considering this parameter for cost–benefit analysis and time-saving. Recent studies have found that the deep learning (DL) model is more popular, since it has a higher prediction accuracy than conventional machine learning (ML) techniques. This study proposes four deep learning approaches: BiLSTM, CNN, GRU, and LSTM models, which is rarely seen in the literature. The model is developed using a large database, including details about cement, fly ash, coarse aggregate, sand, water, age, and blast furnace slag as input variables and compressive strength as an output variable. In this research, 80% of the dataset is used for training, while the remaining 20% is used as a testing dataset. Deep learning models result showed an R-square value of the above around 0.960 at the training phase and almost overhead 0.940 at the testing phase. But GRU model performs better than other models, where the R-square value was the significant level of the overhead of 0.990 at the training phase and also the above almost 0.961 at the testing phase, which is a high-accuracy result for both phases. Thus, this research provides a novel and effective method for predicting HPC's compressive strength, which can help develop sustainable infrastructures without requiring time-consuming and costly experiments.