Keywords: case-based reasoning, analogy-based estimation, building services, capital cost, maintenance costs.

With the advent of Private Finance Initiative (PFI) in the United Kingdom, the need for quality whole life cost (WLC) information has become a pre-requisite for bidding for any such projects. This inevitably has led to a search for new effective methods for estimating whole life cost information. A major difficulty in estimating whole life costing of building services is that initial estimates need to be based on incomplete information on the design. This paper describes a research project, which developed a framework for estimating the whole life-cycle costs of building services based on past data from similar projects. We call this approach Analogy-based Estimation (ABE). Most of the ideas in this work come from the field of Artificial Intelligence (AI). The main reason for us to search for potential solutions in AI was that the nature of the problem we were faced with fitted exactly the types of problems AI is supposed to solve, i.e. problems in domains with incomplete information [1]. The research reported in this paper is a part of an overall approach being developed for whole life costing of building services [2]. A particular problem identified in producing initial estimates of the whole life cost of building services is that the information on the building services design is limited. Analogy Based Estimation (ABE) was investigated as a potential approach to generate more information for estimation purposes and to estimate the capital cost of building services. The performance of ABE was investigated in three test settings. The estimation performance of ABE was compared with the estimation performance of linear regression (LR). ABE outperformed LR on two test settings. However, the estimation performance of ABE was found not to be sufficient for the approach to be applied in industry prior to further investigation. The tests were hindered by lack of data and inconsistencies in the data obtained. Therefore, no conclusions can be made on the applicability of ABE in this particular problem domain and ABE remains a potential approach. The research described in this paper will stand as a starting point for further testing and development of ABE in the estimation of building services. From our studies and observations, it is quite clear that ABE cannot be used for capital estimation of building services in industry based on the data used in this study. The estimation performance of ABE was poor, even though it outperformed LR on two of the three test settings. Based on the test we carried out [2] no firm conclusions can be drawn on the applicability of the technique. However, we did manage to identify the scope for the development of ABE in building services. The data obtained for the project was inadequate and inconsistent. This highlights the need for systematic data collection. The value of high quality data cannot be over-emphasized. An effective estimation tool can only be developed if a sufficient amount of quality data is available. It is expected that the performance of ABE would improve if a separate estimate were generated for each building type and building services system. This is perceived to increase the probability of selecting more appropriate analogies for generating estimates, as there would be less variation in the past solution. Several buildings services system classifications exist. BSRIA is currently promoting a classification [3] as a cost-benchmarking standard for capital costs of building services. High quality data is required to test this proposition.

Some other researchers [4] have compared the performance of ABE, LR and expert estimators. They have demonstrated that expert judgment improves the estimation performance of ABE more than it does that of LR. They suggest that this is because recognizing when an estimate is to be trusted and when disregarded as misleading requires expert judgment, which can be done more readily with ABE than LR. Researchers suggest [4] suggest that there is more value in ABE identifying the closest analogies than in generating the estimates. Furthermore, they also argue that the ultimate test of an estimation tool is to evaluate how much it improves human performance. In their opinion the performance of the estimation tool itself is of secondary interest. Thus, in final conclusion the performance of ABE in building services whole-life cycle cost estimation should be evaluated in combination with expert judgment.

1

Ritter, R., "Logic for Default Reasoning", Artificial Intelligence, 13, 1980.

2

Rintala, K., Kumar, B. and Macleod, I.A., "Analogy-based Estimation in Building Services", Proceedings of International Conference of CIB, W78, IT in Construction in Africa 2001. doi:10.4203/ccp.74.5

3

Nanayakkara, R. and Fitzsimmons, J., "Cost Benchmarks for the Installation of Building Services: Part 3 - A Protocol for Recording Project Costs", BSRIA, Bracknell 1999.

4

Stensrud, E. and Myrtveit, I., "The Added Value of Estimation by Analogy: An Industrial Experiment", Proceedings of the European Software Measurement Conference, Antwerpen, Germany, 549-56, 1998a.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £78 +P&P)