A novel construction cost prediction model using hybrid natural and light gradient boosting

Abstract

In this paper, we compared the predictive capabilities of six different machine learning algorithms – linear regression, artificial neural network, random forest, extreme gradient boosting, light gradient boosting, and natural gradient boosting – and demonstrated that a hybrid light gradient boosting and natural gradient boosting model provides the most desirable construction cost estimates in terms of the accuracy metrics, uncertainty estimates, and training speed. We also present a game theory-based model interpretation technique to evaluate the average marginal contribution of each feature value, across all possible combinations of features, on the model predictions. The comparison between the predicted cost and the actual cost confirms good alignment with R2∼ 0.99, RMSE∼ 0.5, and MBE∼ -0.009. Besides, the proposed hybrid model can provide uncertainty estimates through probabilistic predictions for real-valued outputs. This probabilistic prediction approach produces a holistic probability distribution over the entire outcome space to quantify the uncertainties related to construction cost predictions.