Applying Hybrid Machine Learning for Construction Material Price Prediction and Procurement Cost Optimization

Abstract

Construction material cost is the major component of construction project costs. Among the material cost categories, construction material price fluctuation is the major risk that causes construction cost estimation to be different from actual cost in many countries. In addition, unable to consider the construction material price in construction material procurement cost optimization is uneconomical because may lead to the material being ordered at a period when the price is high. Therefore, a two-staged method for construction material price prediction and a strategic economical construction procurement method is proposed. In the first stage, the Multilayer Perceptron (MLP) is used to predict construction material prices. Then in the second stage, the predicted price of the MLP model was taken as input along with procurement data for the Deep Q Network (DQN) to identify ordering time and quantity at a minimum cost. The application of the proposed method in the Ethiopian construction industry shows that MLP has better performance in predicting cement prices than linear regression. Besides, the DQN algorithm procurement strategy for the nonpolynomial hard problem is < 1% in cost performance than the exact mixed integer linear programming (MILP) method with reasonable solution time. The proposed hybrid model can help construction practitioners to make material-related data-driven decisions.