MINING OF CONCEPTUAL COST ESTIMATION KNOWLEDGE WITH A NEURO FUZZY SYSTEM Wen-der Yu and Yu-ru Lee Institute of Construction Management, Chung Hua University, Taiwan Abstract: Conceptual cost estimation during the early stage of a construction project plays important role for feasibility analysis and project planning. Traditional approaches rely heavily on experienced engineers, and may cause loss of conceptual estimation knowledge of the firm. This paper proposes a method integrates a previous developed conceptual cost estimation method (PIREM) with the ANFIS neuro-fuzzy system for mining of cost estimation data. A case study of residential building projects in Mainland China is conducted to test the proposed method. The testing results show that the proposed method does not only achieve high system accuracy, but also provide many features desirable for estimators such as explicit fuzzy decision rules and graphical sensitivity analysis presentation. Keywords: Conceptual cost estimation; Neuro-fuzzy; Data mining, KDD; China 1. INTRODUCTION The conceptual cost estimation during the engineering planning is important for successful execution of a construction project, since the main structural systems, major construction methods, and most construction materials are determined in that stage. However, due to the lack of detail design information during the planning phase, accurate cost estimation is hard to obtain even for the professional estimators. It was found that the estimators with more estimating experience can do better in this job than who with less. The emerging development of modern artificial intelligence (AI) techniques, such as neuro-fuzzy systems, the aforementioned estimating experience/knowledge can be acquired by learning from historical examples, so that accurate estimation (compared with the detail estimation) could be obtained with very limited available project information. Unfortunately, an essential difficulty is facing the traditional conceptual cost estimation if the knowledge-based approaches are to be adopted—the unit prices of the cost items are variable in the marketplace, so that the estimation knowledge learned previously may not be readily applicable in the future projects. In this paper, the PIREM (Principal items ratio estimation method) [1] approach is integrated with a neuro-fuzzy system, ANFIS [2], to perform data mining (DM) function in the knowledge discovery process of conceptual cost estimation. An application of the proposed DM method for cost estimation of building construction projects in People’s Republic of China (PRC) is selected for demonstration of the proposed method in order to meet the needs caused by more and more construction projects invested by the Taiwan’s businessmen in PRC in the past decade. As the cost estimation system of PRC (so-called “fixed price system [3]”) is different from the cost estimation system in Taiwan (i.e., “bill of quantities”, BOQ system), investors from Taiwan and other countries are unable to obtain accurate conceptual cost estimates for their projects, especially for the first-time investors. In order to conquer the difficulty, knowledge discovery in databases (KDD) [4] techniques are employed to mine the cost estimation knowledge from historical cost data of previous construction projects. The historical cost data are collected from sample projects in the publications published by the Ministry of Construction of PRC [5,6]. Totally, 114 building construction projects were collected and analyzed. The quantities and their unit price information of every cost item are surveyed and calculated separately in PIREM. The ANFIS neuro fuzzy system is adopted to capture the relationships between the influential attributes and the construction cost. The relationships acquired by ANFIS are stored in forms of fuzzy IF-THEN rules, so that the domain experts can visualize and verify the cost estimation knowledge explicitly. With the aid of the proposed approach, the barrier caused by different cost estimation system can be overcome. The testing results show that the cost estimation accuracy can be up to 90.01%, which is considered acceptable during the project conceptual planning phase. The paper is organized in the following manner. In the second section, the previously developed PIREM approach for conceptual cost estimation is reviewed first to provide background knowledge. Then, process for mining of historic cost estimation data is briefly discussed. Fourthly, the ANFIS neuro-fuzzy system as a technique for data mining is introduced. In the fifth section, application of the proposed approach to cost estimation of building