Modeling road construction project cost in the Philippines using the artificial neural network approach

• Main Authors: Roxas, Cheryl Lyne C., Roxas, Nicanor R., Cristobal, Jerald, Hao, Sara Eunice, Rabino, Rachelle Marie G., Revalde, Fulgencio
• Format: text
• Published: Animo Repository 2019
• Subjects: Roads—Design and construction—Estimates > Data processing; Civil Engineering
• Online Access: https://animorepository.dlsu.edu.ph/faculty_research/147; https://animorepository.dlsu.edu.ph/context/faculty_research/article/1146/type/native/viewcontent
• Institution: De La Salle University

Incomplete data and several unforeseen factors affect the accuracy of project cost estimates, especially during the conceptualization stage. When stakeholders need an immediate estimate of the budget for a project, in-depth cost analysis may take time, sacrificing resources for feasibility studies. In the Philippines, a more effective and efficient early cost estimation method is recommended to ensure proper utilization of government funds. In this paper, the artificial neural network technique was adopted to model the local total road project cost. Data collection included 41 road projects with each having 15 factors were recorded, namely: road type, location (region), length of road, duration of project, capacity, pavement thickness, pavement width, shoulder width, earthworks volume, average site clearing/grubbing area, presence of water body, soil conditions, surface class, gross domestic product and consumer price index. After correlation analysis, 7 input variables were finalized. These are the soil condition, surface class, gross domestic product, presence of water body, pavement width, road type and capacity. Several simulations were performed in MATLAB software to determine the best total road project cost model. The best neural network architecture consists of 7 input variables, 12 neurons in the hidden layer and 1 output variable. This neural network model satisfactorily predicted the total cost with coefficient of correlation values of 0.97168, 0.95188, and 0.99036 for training, validation and testing phases, respectively.