A Machine Learning Approach for Conceptual Cost Estimation

Abstract

It is desirable to predict construction cost with a high level of accuracy in the early phase to compare the budgetary with feasibility determinations. Additionally, it is required to be as quick as possible. However, the accuracy of the cost estimation depends on the design details which are extremely limited in such an early phase, rendering numerous uncertainties and dynamics which are hard to control and foretell.

Currently, diverse approaches and techniques are being used and refined to reach the ultimate goal; the cost estimation is to accurately forecast the final cost of a project with no design details available. Generally, estimators’ experience plays a critical role here, and the availability of historical cost data is also crucial. The process is significantly dependent on an export-driven approach. However, decisions made by experts can be subjective and error-prone especially when the relationships between cost drivers and the target cost are not fully understood or even identified. Consequently, cost estimation to a fair level of accuracy is hardly possible to achieve manually within a restricted time. In recent years, civil engineering domain has begun to consider machine learning technique as an optimal approach in tackling the predictive problem through a data-driven approach. Adaptive Network-based Fuzzy Inference System (ANFIS) (a hybrid model of Artificial Neural Network and Fuzzy Inference System) is advantageous in managing uncertainties and representing knowledge. This research aims at investigating the applicability of using the ANFIS for cost estimation during the conceptual phase.