Construction Cost Prediction for Residential Projects Based on Support Vector Regression

Accurate prediction of construction cost with the use of limited information in the initial phase of a construction project is critical to the success of the project. However, traditional cost estimation methods have poor accuracy and efficiency. It is important to utilize the knowledge gained from past projects and historical cost data to predict a new project’s cost. Therefore, this research tries to develop a new methodology based on Support Vector Regression (SVR) for improving the accuracy and efficiency of prediction on a residential project’s total construction cost and its main component costs including engineering project cost, installation project cost and decoration project cost. In this research, we constructed 15 attributes that correspond with the project characteristics and market price fluctuations, and developed 4 SVR models to predict the residential project’s costs. To verify the prediction performance of the proposed model, a case study was performed on 84 residential projects in Chongqing, China. BP Neural Network (BPNN) and Random Forest (RF) were also used to compare the accuracy and stability of prediction results. The results show that the suggested SVR models achieve higher accuracy with 98.32% of the overall cost estimation compared with other models. This research shows that the developed model is effective in early decision making and cost management since the construction cost and its component cost can be predicted accurately before the completion of a project’s design stage.