Post on 03-Feb-2023
1
Quantitative Construction Research:
A Paradigm Shift from Traditional Statistical Techniques to AI?
Goh Bee Hua
Publication details:
Goh, B.H. (2008), “The state of applications of quantitative analysis techniques to construction economics and management (1983 to 2006)”, Construction Management and Economics, Taylor & Francis, 26(5), 485-497.
2
Abstract
The use of artificial intelligence (AI) methods, such as artificial neural networks (ANNs), in the
construction field has been in experimentation since the evolution of one of the most important ANN
paradigm known as Backpropagation in the mid-eighties. This paper presents a review of studies that
have applied biological-based AI methods, such as genetic algorithms (GAs) and ANNs, to the
simulation and optimization of construction processes, and the modelling and forecasting of
construction variables such as tender bids, demand and cost, respectively. Some of the highlighted
works include the use of ANNs and GAs to solve complex construction site operational problems; to
identify the most appropriate construction method, plant combination and labour allocation, so as to
optimize costs, time and production; to examine the relationships between variables in the
construction bidding process for the purpose of predicting future tender bids; to model the complex
nonlinear relationship between demand for construction and its influencing economic factors; and to
model and estimate construction cost using subjective or 'noisy' data. The key benefits cited include a
more efficient method of simulating construction activity owing to the parallel processing of
information; a more effective search and optimization approach that can quickly locate high
performance regions in extremely large and complex search spaces; the building of more complex
nonlinear models to produce realistic and accurate forecasts; and the availability of techniques that
can perform nonlinear modelling and adaptation automatically without the need for functional
assumptions. For quantitative research in the field of construction management and economics, the
review serves to formalise the transition from traditional statistical techniques to those that are AI-
based. It is timely to acknowledge a paradigm shift in quantitative construction research so that
researchers can better prepare themselves for the future by acquiring new skills and knowledge.
Keywords: artificial neural networks, statistical techniques, quantitative research,
construction management, construction economics
3
Introduction
The construction industry is most often associated with complexity; in terms of the nature of work,
management of processes, organization of parties’ relationships, and including its inter-relationship
with the rest of the economy. In dealing with such complexities, conventional decision-analysis
techniques were regarded as less acceptable by practicing construction personnel, especially when
they involve advanced mathematics such as probabilistic methods and multi-attribute utility theory
(Ahmad and Minkarah, 1990). Levitt (1987) highlighted that the key requirements for the resolution
of construction problems, as well as the performance of construction management tasks are expert
knowledge, judgment and experience. Many expert system applications in construction-related areas
have been made on this basis and the intention is for the system to attempt to model the intelligent
reasoning and problem-solving capabilities of the human brain. However, as neural networks
research started to flourish in the mid eighties, it was realized that expert systems lack the ability to
learn by themselves, generalize solutions and adequately respond to highly correlated, noisy,
incomplete or previously unseen data (Wassermann, 1989; Pao, 1989). In contrast, neural networks
have these abilities by attempting to model the brain learning, thinking, storage and retrieval of
information, as well as associative recognition (Moselhi et al., 1991). Successful applications of
neural networks in the field of construction management and economics have stimulated further
interest in exploring the capabilities of other biological-based methods such as genetic algorithms.
These are parallel, adaptive search algorithms inspired by the mechanisms of biological evolution,
and have been used to solve a variety of optimization problems (Goldberg, 1989). So far, they have
either been used alone to solve construction optimization problems or in combination with neural
networks to provide a synergy with more problem-solving power. In essence, neural networks mimic
brain functions while genetic algorithms mimic the process of evolution and they both have a
common motivation, that is, to learn from nature’s efficiency.
This paper presupposes a paradigm shift from the use of traditional statistical approaches to
biological-based AI techniques in construction management and economics research. Hence, it is
seen to be timely to present a review of past and recent studies in order to formalize this transition so
that researchers can better prepare themselves for the future by acquiring new skills and knowledge.
The paper focuses on two key areas of construction management and economics research where the
paradigm shift has been most predominant. They are in the areas of modelling and forecasting of
construction variables such as tender price, construction demand and cost, and the simulation and
optimization of construction processes. It also serves the purpose of narrowing down the scope of the
review as the field of construction management and economic is inherently broad and diverse.
4
Following the review, the causes for the shift are discussed and the general performance of these
biological-based AI methods is evaluated to see if the transition has indeed made positive impact on
this field of research.
Modelling construction economics variables using traditional approaches
In the construction industry, regression techniques have often been used to model and forecast
construction variables owing to their relative simplicity in both concept and application. In essence,
regression techniques are concerned with the modelling of relationships among variables and they
quantify how a response variable is related to a set of explanatory variables. Univariate time-series
methods are also popular, especially when there is lack of information on such explanatory variables,
and the forecast has to rely on identifying and extrapolating the underlying pattern of the historical
data from the series itself. In essence, time-series forecasting treats the system as a black box and
makes no attempt to discover the factors affecting its behaviour. In support of Taylor and Bowen’s
(1987) suggestion, Wang and Mei (1998) acknowledge that there are two categories of quantitative
forecasting: (1) the causal method and (2) the time-series method.
In the domain of tender bid and price forecasting, both regression and time-series approaches have
been applied. Carr and Sandahl (1978) applied multiple regression analysis to competitive bidding to
provide decision support on whether to bid or not to bid and to estimate the bid markup. Flanagan
and Norman (1983) used simple linear and curvilinear regression analyses to establish the relationship
between quantity surveyor’s price prediction made at the design stage and the low bids received from
contractors. While McCaffer et al. (1983) noted that higher levels of accuracy can be achieved for
tender price forecasting using time-series methods, Taylor and Bowen (1987) adopted a hierarchy of
time-series forecasting methods, ranging from simple moving average and single exponential
smoothing to advanced decomposition and Box-Jenkins techniques, to forecast future building price
index levels. However, Runeson (1988) still relied on multiple regression analysis to quantify the
effects on the price level of variations in the degree of competitiveness in the industry, using Market
Conditions Index as the dependent variable. Fellows (1987 and 1991) used the Box-Jenkins
methodology, in particular, to predict building cost and tender price indices with the objective of
managing escalation by forecasting the effects of inflation in the UK building industry. Adopting the
same approach, Goh (1995) examined the effects of inflation on public industrial projects in
Singapore via tender price index forecasting.
5
Similarly, these conventional approaches have been used to model and forecast contractors’ workload
or, in broader terms, demand for construction. A comparative study was undertaken to evaluate the
forecasting accuracy of the regression and time-series methods, using the Box-Jenkins technique as
the benchmark (Oshobajo and Fellows, 1989). Other works relied mainly on regression analyses,
such as, Killingsworth Jr. (1990) developing a demand forecasting model based on a regression
analysis of the leading indicators of industrial construction in the US, and Tang et al. (1990) applying
regression analyses to project values of total construction activities for the Thai construction industry.
In addition, models of UK private-sector quarterly construction demand were developed by Akintoye
and Skitmore (1994) using a priori selected leading indicators and applying the regression technique
with lagged terms. More recently, Goh (1998) compared the accuracy of the Box-Jenkins and the
multiple regression techniques in forecasting residential construction demand in Singapore. Owing to
the extensive use of regression analyses in demand forecasting, Goh (1999) extended her study to
explore the accuracy of both linear and nonlinear regression methods applied to modelling
Singapore’s residential, industrial and commercial construction demand.
For construction cost estimation and prediction, the conventional approaches adopted have also not
departed from causal and time-series methods. As a cost estimating tool, McCaffer (1975) illustrated
with examples the general use of regression analysis. Specifically for reinforced concrete framed
structures, Bowen (1982) examined the feasibility of econometric cost modelling as an alternative
estimating approach. On the prediction of construction costs for public school buildings in Jordan,
Al-Momani (1996) developed an empirical model by experimenting the use of linear and loglinear
regression methods. For assessing the variations in construction labour and material costs, Wang and
Mei (1998) presented a new approach, based on time-series forecasting concepts, to predict
construction cost indices in Taiwan.
Modelling construction management variables using traditional approaches
In the construction management process, the key consideration is to use available resources in the most
efficient manner in order to minimise time and cost with the ultimate goal of enhancing productivity and
cash-flow of a project. Past studies have ranged from the application of deterministic critical path
methods such as CPM and PERT to the adoption of stochastic or probabilistic approaches such as
Monte Carlo simulation and related optimisation techniques. Cusack (1984) outlined the historical
development of mathematical models in the planning and control of construction projects and discussed
the limitations of such deterministic techniques. He stressed the need to model and inter-relate time cost
parameters in a dynamic way through simulation and optimisation. In concurrence, Bennett and
6
Ormerod (1984) re-emphasised the claim that the introduction of simulation methods for construction
management is likely to have as great an impact on the construction industry as did the introduction of
network planning and scheduling methods some two decades ago. In essence, simulation is a technique
which is employed to form a management tool in an uncertain environment in order to augment the
decision maker’s intuition and experience. Stochastic simulation typically generates durations and costs
for each activity in a plan by randomly calculating a feasible value for each from a statistical probability
frequency distribution which represents the range and pattern of possible outcomes for an activity. With
the use of computer-based simulation, studies can be undertaken to determine the most appropriate
construction method, plant combination and labour allocation, so as to optimise costs, time and
production, and to predict probable project durations, costs and production for estimating and tendering
purposes (Flood, 1990).
Owing to the diversity of construction work and range of methods and resources employed, computer
simulation and optimisation methods have dominated research studies for solving complex construction
operational problems. However, based on mathematical formulation of optimum rates the concreting
and formwork operation, Gates and Scarpa (1978) derived formulas to optimise the rate of pour that
dictates formwork design, the number of uses of formwork, and the number and height of lifts.
Customised simulation tools have also been developed for specific construction operations such as the
SCRAPESIM for earth-moving owing to the cost significance of this activity where inaccuracies in
estimating may greatly influence the profit margin of an earth-moving contractor (Clemmens and
Willenbrock, 1978). For solving large-scale operational problems, interactive man-computer graphics
systems and heuristic modelling approaches for simulation have been proposed (Paulson Jr., 1978).
Different applications of stochastic techniques to simulate construction activity have been described by
Pilcher and Flood (1984) and their review of previous and current research serves to acknowledge the
potentially usefulness of computer-based simulation for construction managers in supplementing
intuition and expertise. To further enhance existing simulation tools for construction operations,
Paulson et al. (1987) proposed a system that combines videotape data collection from field construction
operations, statistical reduction and analysis of data, and computer-based simulation modelling. More
recent research (Alkoc and Erbatur, 1998) extends the boundary of simulation models to include a
sensitivity analysis of the concreting operations to a set of possible resource combinations.
For construction resource estimation, mainly dealing with time and cost, the use of stochastic simulation
techniques has been prevalent. Carr (1979) presented a simulation model for uncertainty determination
of the timing of each construction activity, even when the activity durations are not independent of each
other. As an integrated approach, Woolery and Crandall (1983) proposed the use of a stochastic
7
network model, based on Monte-Carlo simulation, to model the dependencies between network
activities for construction scheduling. As a means to augment pre-contract planning for large and
complex projects or for re-planning work in progress, the use of Monte-Carlo simulation was also
proposed for developing construction resource models (Baxendale, 1984). A suite of computer
programmes, known as the Construction Project Simulator, was developed to generate simulation
outputs for duration and cost predictions, and cash flow curves, based on costs, resources, weather and
productivity data (Bennett and Ormerod, 1984). Considering mathematical approaches as complex
solutions to construction resource modelling, Cusack (1985) presented a simplified approach to the
planning and control of cost and project duration by applying heuristics to model the time-cost
relationship. As a major addition to the repertoire of existing computer-based project scheduling tools,
Ahuja and Nandakumar (1985) developed a comprehensive model to simulate and combine the impact
of the significant uncertainty variables, such as weather, space congestion, workmen absenteeism, and
incorporates it in the activity duration estimates as it revises them. Further research developments in the
area of project scheduling include the development of a systematic methodology for stochastic
scheduling involving eight steps (Karni, 1987) and the formulation of a linear integer model for
optimisation of project schedules to achieve the primary objective of minimising total project cost
(Karshenas and Haber, 1990).
Research studies on the estimation and prediction of construction cash flow typically dealt with
generating alternative forms of the S-curve. The techniques applied ranged from mathematical
formulation of the S-curve to stochastic simulation of the patterns of construction project expenditure.
As a general approach, Tucker (1986) introduced the mathematical formulation of construction cash
flow curves using an analogy between the probability of failure in reliability theory and the probability
of payment during construction. It permits the exploration of any function to define the payment
completion rate from which the cumulative payment and payment density functions can be derived.
More specifically for the post hoc examination of construction project net cash flows, Kenley and
Wilson (1989) proposed a model based upon the logit transformation which was found to be very
flexible and capable of adapting to the wide degree of inter-project variability. In contrast with
deterministic approaches, simulation of expenditure patterns of construction projects to analyse and
examine their shapes was proposed by Khosrowshahi (1991) as a means of facilitating mathematical
models for ease of application, user involvement and user comprehension. To further improve on
current standard value S-curves, Kaka and Price (1991) focused their attention on cost commitment
curves instead, especially to help contractors generate the value and cash-out curves at the tendering
stage. The applicability of such curves was tested on several project classification criteria and curve-
fitting was carried out using the logit transformation technique (Kaka and Price, 1993).
8
Similarly, the types of techniques applied in construction productivity studies ranged from deterministic
modelling to stochastic simulation. An interactive system for the analysis of construction operations by
integrating quantity development and process simulation to estimate productivity and cost was proposed
by Tavakoli (1985). By applying the learning curve theory to construction productivity, five basic
mathematical models for learning curves, ranging from straight-line to exponential forms, were
evaluated on estimating and predictive abilities (Thomas et al., 1986). Thomas and Yiakoamis (1987)
developed a factor model using multiple regression techniques to mathematically explain variability
caused by the effects of temperature and relative humidity in the daily productivity data.
Equipment selection is also an important consideration in construction management and research studies
often apply simulation and optimisation techniques to experiment different combinations of
construction plant and equipment to improve site productivity or select the least cost option. Woods and
Harris (1980) developed a truck allocation model for concrete distribution by applying computer-based
simulation to obtain the most suitable combination of trucks in the fleet in order to minimise operating
costs and reduce waste space from part loads. However, mathematical techniques were relied upon to
propose an algorithm which can define the least expensive cranage cost for a project based on varying
the type and number of cranes used during construction (Furusaka and Gray, 1984). Both queuing
theory and simulation techniques were adopted by Touran and Taher (1988) to develop a model that can
predict the productivity, as well as determine the optimum fleet size using sensitivity analysis, of the
earth-moving operation. For materials handling, Wijesundera and Harris (1989) developed a dynamic
interactive simulation model to allow for varying crane type, size and location, skip size, delivery
system and construction crew size to evaluate the effects on utilisation levels and costs for different
methods of working.
Biological-based AI applications in construction management and economics
The use of AI, namely expert systems, in the field of construction management and economics began in
the early eighties. Brandon (1990) reviewed the historical development of this AI application in the
construction industry, highlighting its potential and limitations. With knowledge acquisition being the
major bottleneck in the development of expert systems (Bowen and Erwin, 1990), other biological-
based applications started to evolve in this field. Since the early nineties, artificial neural networks
(ANNs) have become popular, especially in the area of construction management. Their superiority
over expert systems stems from the nature of construction problems which is of pattern recognition
rather than deep reasoning about the problem elements (Moselhi et al., 1991). More recently, genetic
9
algorithms (GAs), another biological-based method, are gaining recognition in solving construction
optimisation problems either alone or in combination with ANNs.
In the complex decision-making process of competitive bidding, ANNs have been applied to bid
prediction and analysis, and mark-up estimation. Gaarslev (1991) developed a neural network to predict
the outcome of a future bidding situation based on historical data such as number of bidders, estimated
cost, percentage of work not sub-contracted and project duration. Although McKim (1993) adopted the
same set of data for his study, he extended his objective to include a comparison of the predictive
capabilities of a traditional statistical methods and ANNs. To aid contractors in preparing competitive
bids for building projects, Moselhi et al. (1993) developed a decision-support system which uses ANNs
for mark-up estimation and the Monte Carlo simulation technique for assessing the contractor’s
assessment of project risks. Focusing only on optimum mark-up estimation, Hegazy and Moselhi
(1994) proposed an analogy-based solution, which combines ANNs and GAs, to mark-up estimation
problem, while Li (1996) compared the performance of ANNs to a regression-based method and
identified the effect of different ANN configurations, such as the number of hidden layers and nodes in
each hidden layer, on estimating accuracy. Further improvements on the ANN methodology have been
undertaken on construction mark-up estimation (Li et al., 1999).
For modelling demand for construction, comparative studies have been carried out to evaluate the
forecasting accuracy of ANNs against traditional techniques such as regression and time-series analyses.
Adopting linear and nonlinear forms, Goh (1996) developed two residential construction demand
models using multiple linear regression and the backpropagation ANN, respectively. Further
developments of this comparison include an evaluation of the backpropagation ANN against multiple
loglinear regression and the univariate Box-Jenkins approach (Goh, 1998), and against the performance
of combining ANNs and GAs (Goh, 2000).
For solving complex construction operational problems, the use of ANNs has also been explored
(Flood, 1990a). An ANN approach was proposed by Flood (1989) to achieve an optimal sequencing of
construction tasks with the aim of minimising production time. Based on the efficient solutions
obtained in the study to the sequencing problem, he concluded on the possibility of applying ANNs to
other types of operational problems such as resource allocation, material cutting and site layout. Flood
(1990b) also illustrated with the use of a conceptual ANN model to simulate a construction process by
linking network modules, representing a queuing facility, storage facility or productive resource, to
design a complete network description.
10
As time and cost are two important inter-related factors to be planned and controlled in construction
projects, the capabilities of both ANNs and GAs have been explored. A judgment-based forecasting
approach, using multiple regression techniques and ANNs, was proposed by Al-Tabtabai et al. (1997) to
identify schedule variances from a baseline plan for typical construction projects. It essentially uses
ANNs to capture the complex nonlinear decision-making process of project experts involved in
schedule monitoring and prediction. For time-cost optimisation problems in construction planning, Li
and Love (1997) evaluated the performance of a basic GA system and implement modifications to the
crossover and mutation operations. Besides outperforming the basic system, the improved GA system
was found to be able to generate a whole class of alternative solutions close to the optimum. On the
other hand, Adeli and Karim (1997) adopted ANNs for solving the nonlinear optimisation problem for
construction project scheduling. By varying the construction duration, one can solve the cost-duration
trade-off problem, and obtain the global optimum schedule and the corresponding minimum
construction cost.
Similarly for studies on construction budget performance and cost estimation, both ANNs and GAs
have been applied. A neural network approach was used to identify the key management factors that
affect budget performance in a construction project and develop a prediction model based on their
complex relationships (Chua et al., 1997). Adeli and Wu (1998) formulated the regularisation neural
network for construction cost estimation by incorporating a regularisation term in the error function to
compensate for the overfitting problem and improve estimation outside the scope of available data
points. The performance of this new computational model depends only on the training examples; and
not on the architecture of the network, learning parameters and number of training iterations. Hegazy
and Ayed (1998) also used ANNs for their study on parametric cost estimation but relied on simplex
optimisation and GAs to determine the network weights. Accordingly, the weights that produced the
best cost prediction for the historical cases were used to find the optimum network. For industry-level
cost prediction, William (1994) developed a backpropagation ANN model to predict the changes in
construction cost indexes using related economic indicators as model inputs. The performance of the
ANN model was compared with the exponential smoothing and linear regression models.
In construction productivity studies, ANNs have been used to estimate operation productivity, as well as
labour productivity. Chao and Skibniewski (1994) adopted a neural-network and observation-data-
based approach to estimate the production rate for the excavation and hauling operation. Experimenting
on a different construction operation, Portas and AbouRizk (1997) developed a three-layered ANN with
a fuzzy output structure to estimate construction productivity for concrete formwork tasks. For
modelling construction labour productivity, Sonmez and Rowings (1998) adopted a methodology based
11
on regression and ANNs techniques to develop prediction models for concrete pouring, formwork and
concrete finishing tasks. It has presented an approach for the evaluation of the impact of multiple
factors, considered simultaneously, on labour productivity.
For the selection of construction equipment, studies have adopted GAs and ANNs to search for an
optimum equipment selection, and classify new and obsolete equipment based on their production rates,
respectively. To classify earthmoving equipment according to the speed of operation, Karshenas and
Feng (1992) used ANNs as a modular approach to facilitate the inclusion and removal of new and
obsolete equipment considered by the network. On the other hand, for the optimum selection of
excavating and haulage equipment in opencast mining, Haidar et al. (1999) designed a decision-support
system using a hybrid knowledge-based system and GAs. The knowledge base relates mainly to the
selection of equipment in broad categories while the advanced GAs search techniques find the input
variables that can achieve the optimal cost.
Discussion on the paradigm shift
The contentious issue to be decided here is whether there has been a paradigm shift. In order to explore
this issue further, two vital questions have to be addressed. They are: (1) if there has been no paradigm
shift, is there a need for one?; and (2) if a paradigm shift has taken place, is it effective in terms of
having improvements over traditional methods based on the existing paradigm?
The review of studies having applied traditional and biological-based AI techniques has been
summarised in Tables 1 and 2 for variables relating to construction economics and construction
management, respectively. Some studies had carried out a comparison of the performance of traditional
and biological-based AI techniques and the results have also been included. It is obvious that most of
the reviewed studies, i.e. 4 out of 5, have found AI techniques to be more accurate for the construction
economics forecasting studies. The superiority of non-traditional techniques over conventional ones has
been acknowledged in some of the construction management studies, even though no formal
comparison had been undertaken. This simple analysis can help to shed some light on the two
fundamental questions to be addressed here.
Firstly, if there has been no paradigm shift, is there a need for one? Bowen and Edwards (1985, p. 203)
believe that “pressure to adopt a new paradigm presupposes some failure of or dissatisfaction with the
existing one”. Based on the review, it is noted that there has been an increasing number of studies
experimenting with biological-based AI techniques in the field of construction economics and
12
management. Hence, can this be regarded as a signal for at least some failure or dissatisfaction with
conventional techniques based on the existing paradigm. The comparative studies have specifically
used accuracy as the performance criterion for evaluating the modelling techniques. Does this mean
that the existing paradigm has been unable to fulfil the basic requirement of a quantitative model, that is,
to generate accurate forecasts?
Secondly, if a paradigm shift has taken place, is it effective in terms of having improvements over
traditional methods? The obvious answer for accuracy improvements is ‘yes’ for studies on modelling
construction economics variables. Time improvement has been cited in one of the construction
management studies (Flood, 1990b). The common explanation provided by many researchers on the
success of ANNs has been their inherent ability to learn and generalise from previous examples, and
they are particularly suitable for solving nonlinear and multivariate problems commonly found in the
domain of construction economics and management. Traditional methods such as regression and
simulation would require researchers to ascribe functional forms to fit the data patterns and specify
probability distributions for the variables concerned, respectively. Very often, these tasks are carried out
with little technical knowledge on the part of the researcher and judgmental input is heavily relied upon.
On the contrary, they could have been automatically handled using ANNs and GAs, eliminating the
element of subjectivity. Hence, in this case, is the subjectivity element inherent in conventional
methods the cause for the hypothesized failure or dissatisfaction with techniques based on the existing
paradigm?
Model specification has always been a problem in econometric modelling. In order to avoid
encountering the problem of having to ascribe a functional form to complex non-linear relationships,
the general research strategy has been to formulate a provisional model on the basis of economic
theory, experience and intuition, and then locate suitable data and fit the model. Dougherty (1992)
criticised this approach, explaining that the final version of the model would be satisfactory because
its specification would have been skilfully massaged to fit the particular data set, not that it really
corresponds to the true model. Besides, such judgmental interventions of the modeller would have
introduced personal biasness, further undermining the validity of such models.
However, the use of regression techniques in econometrics allows causal analysis to be carried out to
explain the relationships between dependent and independent variables. The coefficient estimates and
the various established significance tests allow the extent of the relationships to be clearly defined. In
contrast, ANNs lack this capability as they are not able to assess the relative importance of the input
factors used by the model to arrive at its conclusion. The absence of this explanatory power in ANNs
13
has traditionally classified them as "black boxes" whose complex inner workings somehow transform
inputs into predicted outputs. Traditionally, they are not designed to exploit existing expertise;
instead they serve as highly intensive computational machines, deriving their ability to forecast
through a long learning process based on pre-defined sets of input-output values.
In the search for a new paradigm in cost modelling and price forecasting, Bowen and Edwards (1985, p.
207) commented that “what is needed is an approach which will harness the strengths and minimise the
weaknesses of all that has been developed to date, if the wasteful step of discarding the work of the last
fifteen years is to be avoided.” Similarly, if the use of AI techniques is the way forward for construction
management and economics modelling research and, hence, acknowledged as the new paradigm, the
inherent shortcomings of such methods need to be looked into carefully. Apparently, work in this
direction has begun since the early nineties when Garson (1991) developed a simple and intuitive
method for interpreting the interconnection weights in order to understand the impact various input
processing elements have on the output processing element as a means to eliminate the black-box
image of ANNs. More recently, work has been done to extract rules from neural networks (Fu,
1994). The main intention is to formalize the relationship between a neural network and a rule-based
system. On this basis, several studies have been undertaken to support construction mark-up
estimation (Hegazy and Moselhi, 1994; Li, 1996) as it is often regarded to be the most important and
difficult decision for a contractor when making a bid for a project since the mark-up estimate has to
be low enough to win the contract but high enough to make a profit. Further development includes
investigating the use of a rule-extraction method, known as KT-1, to automatically formulate rules for
the trained ANN instead of having to craft the rules manually (Li et al., 1999).
Bowen and Edwards (1985, p. 200) also suggest that “the need for a new paradigm is perceived as being
entirely an academic one”, that is to say, “pressure to consider a new paradigm emanates almost entirely
from an academic pursuit of knowledge.” They were in the view that although academics may see the
need as achieving a significant milestone, it holds little interest for practitioners. On this note, it would
be useful to highlight that the rapidly growing interdisciplinary field of nonlinear modelling and
forecasting has been due largely to the increasing availability of computer resources which allow for the
collection of increasingly large data sets and the analysis of data sets with numerically intensive
algorithms. Casdagli and Eubank (1991) attributed the growth in the field to the increasing recognition
of the ubiquity and importance of the effects of nonlinear dynamics in the natural and social sciences.
Among a wide range of disciplines working on different aspects of nonlinear modelling, advancement
in AI techniques has been one of the greatest. Hence, if it was true that the need for a shift had arisen
from an academic pursuit of knowledge, then it would have to be seen as a step in the positive direction
14
of exploring the use of more computationally rigorous techniques that can model real-world situations
more exactly. With regard to modelling techniques that rely on manual specification of functional
forms, Newbold and Bos (1990) stressed that it would be quite unrealistic to hope that any simple model
and associated assumptions can provide a perfectly accurate description of the complexity that exists in
the real world. On the basis of testing out state-of-the-art methods, it is natural for research and
development work to stem from the academic arena with no significant role for practitioners to play at
this developmental phase. The anticipated challenge in the future would be to convince the industry on
the viability of such biological-based AI modelling techniques, in terms of their ability to generate good
quality forecasts with time and cost savings. In this way, the ideal link between academic institutions
and the industry can be further established with sustainability.
Table 1. A summary of studies on modelling construction economics variables
Construction Economics Variables
Conventional Techniques Applied
Biological-based Techniques Applied
Remarks
Tender Bids / Price
Carr & Sandahl (1978) Flanagan & Norman (1983) McCaffer et al. (1983) Taylor & Bowen (1987) Fellows (1987) Runeson (1988) Fellows (1991)
Gaarslev (1991) McKim (1993) * Moselhi et al. (1993) Hegazy & Moselhi (1994) Li (1996) * Li et al. (1999)
Comparative studies (*) have found ANNs to be superior to regression analysis in terms of forecasting accuracy.
Demand
Oshobajo & Fellows (1989) Tang et al. (1990) Killingsworth Jr. (1990) Akintoye & Skitmore (1994) Goh (1999)
Goh (1996) * Goh (1998) * Goh (2000)
Comparative studies (*) have found ANNs to be superior to multiple linear and loglinear regression, and comparable to the Box-Jenkins approach, which is a univariate time-series technique, in terms of forecasting accuracy.
Cost
McCaffer (1975) Bowen (1982) Al-Momani (1996) Wang & Mei (1998)
Williams (1994) *
The comparative study (*) has shown that ANNs did not outperform the exponential smoothing and simple linear regression techniques, in terms of forecasting accuracy.
15
Table 2. A summary of studies on modelling construction management variables
Construction Management Variables
Conventional Techniques Applied
Biological-based Techniques Applied
Remarks
Operations
Gates & Scarpa (1978) Paulson Jr. (1978) Clemmens & Willenbrock (1978) Pilcher & Flood (1984) Paulson et al. (1987) Alkoc & Erbatur (1998)
Flood (1989) Flood (1990a) Flood (1990b) *
No formal comparative studies have been conducted but it has been said in a study (*) that ANNs have dramatic improvement over conventional computer simulation implementations in terms of the amount of simulation time incurred.
Resources (in relation to time & cost)
Carr (1979) Woolery & Crandall (1983) Baxendale (1984) Bennett & Ormerod (1984) Cusack (1985) Ahuja & Nandakumar (1985) Karni (1987) Karshenas & Haber (1990)
Al-Tabtabai et al. (1997) * Li & Love (1997) Adeli & Karim (1997)
No formal comparative studies have been conducted but it has been acknowledged in a study (*) that multiple regression is suitable for modelling linear relationships while ANNs are more flexible in capturing non-linearity.
Project Budget / Cash-flow
Tucker (1986) Kenley & Wilson (1989) Khosrowshahi (1991) Kaka & Price (1991) Kaka & Price (1993)
Chua et al. (1997) * Adeli & Wu (1998) Hegazy & Ayed (1998)
No formal comparative studies have been conducted but it has been highlighted in a study (*) that the ANN model has the ability to give good predictions even with previously unseen data and incomplete information.
Productivity
Tavakoli (1985) Thomas et al. (1986) Thomas & Yiakoamis (1987)
Chao & Skibniewski (1994) Portas & AbouRizk (1997) Sonmez & Rowings (1998) *
No formal comparative studies have been conducted but a study (*) has applied multiple regression and ANNs where the former was used to select significant inputs for the latter. The study has concluded that ANNs have shown potential for quantitative evaluation of the effects of multiple factors on productivity, especially when interactions and nonlinear relations are present.
Equipment Selection
Woods & Harris (1980) Furusaka & Gray (1984) Touran & Taher (1988) Wijesundera & Harris (1989)
Karshenas & Feng (1992) Haidar et al. (1999)
No comparative studies have been conducted.
16
Conclusions
Two vital questions have been addressed in the discussion. They are:
1) If there has been no paradigm shift, is there a need for one?
The review has revealed that there has been an increasing number of studies experimenting with
biological-based AI techniques in the field of construction economics and management. This could be
regarded as a signal for at least some failure or dissatisfaction with conventional techniques based on the
existing paradigm. Failure of the existing paradigm may be due to the inability of traditional techniques
to produce quantitative models that fulfil the basic requirement of generating accurate forecasts.
2) If a paradigm shift has taken place, is it effective in terms of having improvements over traditional
methods based on the existing paradigm?
The review has shown that, in terms of accuracy, improvements have been acknowledged in a large
proportion of construction economics studies. Time improvement has been cited in one of the
construction management studies. It has been questioned if subjectivity inherent in conventional
methods, in terms of requiring researchers to ascribe functional forms or specify probability
distributions to data patterns or modelling variables respectively, the cause for the hypothesized failure
or dissatisfaction with techniques based on the existing paradigm.
This has prompted a general discussion on the strengths and weaknesses of conventional and state-of-
the-art techniques. Based on the rationale that “what is needed is an approach which will harness the
strengths and minimise the weaknesses of all that has been developed to date, if the wasteful step of
discarding the work of the last fifteen years is to be avoided”, it has been concluded that the way
forward for construction management and economics modelling research is to explore how biological-
based AI techniques could be further enhanced to incorporate explanatory capability. A formalized
relationship between a neural network and a rule-based system needs to be established for future
research.
On the issue if “the need for a new paradigm is perceived as being entirely an academic one”, it has
been concluded that even if it is, it would have been in the right direction of exploring, as well as
exploiting, the use of more computationally rigorous techniques that can model real-world situations
more accurately. It has been concluded that the ideal link between academic institutions and the
industry could then be established, with sustainability, if research and development work could lead to
17
helping the industry adopt techniques that bring about time and cost savings, and in this case, biological-
based AI models that produce good quality forecasts.
References
Adeli, H. and Karim, A. (1997) Scheduling/cost optimization and neural dynamics model for
construction, Construction Engineering and Management, 123(4), 450-58.
Adeli, H. and Wu, M. (1998) Regularization neural network for construction cost estimation,
Construction Engineering and Management, 124(1), 18-24.
Ahmad, I. and Minkarah, I. (1990) Decision analysis and expert system technology: a construction
industry perspective, in Proceedings of the International Symposium on Building Economics and
Construction Management, CIB, Sydney, Australia, 363-72.
Ahuja, H.N. and Nandakumar, V. (1985) Simulation model to forecast project completion time,
Construction Engineering and Management, 111(4), 325-42.
Akintoye, A. and Skitmore, M. (1994) Models of UK private sector quarterly construction demand,
Construction Management and Economics, 12(1), 3-13.
Alkoc, E. and Erbatur, F. (1998) Simulation in concreting operations: a comparison of models and
resource combinations, Engineering, Construction and Architectural Management, 5(2), 159-73.
Al-Momani, A.H. (1996) Construction cost prediction for public school buildings in Jordan,
Construction Management and Economics, 14(4), 311-17.
Al-Tabtabai, H., Kartam, N., Flood, I. and Alex, A.P. (1997) Expert judgment in forecasting
construction project completion, Engineering, Construction and Architectural Management, 4(4), 271-
93.
Baxendale, T. (1984) Construction resource models by Monte Carlo simulation, Construction
Management and Economics, 2(3), 201-17.
Bennett, J. and Ormerod, R.N. (1984) Simulation applied to construction projects, Construction
Management and Economics, 2(3), 225-63.
Bowen, P.A. (1982) An alternative estimating approach, Chartered Quantity Surveyor, February, 191-4.
Bowen, P.A. and Edwards, P.J. (1985) Cost modelling and price forecasting: practice and theory in
perspective, Construction Management and Economics, 3(3), 199-215.
Bowen, P.A. and Erwin, G.J. (1990) The impact of second generation expert systems on building design
evaluation, in Proceedings of the International Symposium on Building Economics and Construction
Management, CIB, Sydney, Australia, 419-27.
Brandon, P.S. (1990) Expert systems – after the hype is over, in Proceedings of the International
Symposium on Building Economics and Construction Management, CIB, Sydney, Australia, 314-44.
18
Carr, R.I. (1979) Simulation of construction project duration, ASCE Journal of the Construction
Division, 105(CO2), June, 117-28.
Carr, R.I. and Sandahl, J.W. (1978) Bidding strategy using multiple regression, ASCE Journal of the
Construction Division, 104(CO1), March, 15-26.
Casdagli, M. and Eubank, S. (1991) in Proceedings of the Workshop on Nonlinear Modelling and
Forecasting, Santa Fe, New Mexico, September, Vol.12, pp. xiii-xix.
Chao, L.C. and Skibniewski, M.J. (1994) Estimating construction productivity: neural-network-based
approach, Computing in Civil Engineering, 8(2), 234-51.
Chua, D.K.H., Kog, Y.C., Loh, P.K. and Jaselskis, E.J. (1997) Model for construction budget
performance – neural network approach, Construction Engineering and Management, 123(3), 214-22.
Clemmens, J.P. and Willenbrock, J.H. (1978) The SCRAPESIM computer simulation, ASCE Journal of
the Construction Division, 104(CO4), December, 419-35.
Cusack, M.M. (1985) A simplified approach to the planning and control of cost and project duration,
Construction Management and Economics, 3(3), 183-98.
Dougherty, C. (1992) Introduction to econometrics, Oxford University Press, Oxford.
Furusaka, S. and Gray, C. (1984) A model for the selection of the optimum crane for construction sites,
Construction Management and Economics, 2(2), 157-76.
Fellows, R.F. (1987) Towards the prediction of escalation on UK building projects, in Proceedings of
the CIB W-55 Symposium, Copenhagen, Vol. B, 57-66.
Fellows, R.F. (1991) Escalation management: forecasting the effects of inflation on building projects,
Construction Management and Economics, 9(2), 167-84.
Flanagan, R. and Norman, G. (1983) The accuracy and monitoring of quantity surveyors’ price
forecasting for building work, Construction Management and Economics, 1(2), 157-80.
Flood, I. (1989) A neural network approach to the sequencing of construction tasks, in Proceedings of
the 6th International Symposium on Automation and Robotics in Construction, San Francisco, CA, 204-
11.
Flood, I. (1990a) Solving construction operational problems using artificial neural networks and
simulated evolution, in Proceedings of the International Symposium on Building Economics and
Construction Management, CIB, Sydney, Australia, 197-208.
Flood, I. (1990b) Simulating the construction process using neural networks, in Proceedings of the 7th
International Symposium on Automation and Robotics in Construction, Bristol, UK, 374-82.
Fu, L.M. (1994) Rule generation from neural networks, IEEE Transactions on Systems, Man and
Cybernetics, 24(8), 1114-24.
Gaarslev, A. (1991) Neural networks- technique tested on classical bidding strategic data set, in
Proceedings of the Symposium on Management, Quality and Economics in Buildings, CIB W-65,
Lisbon, Portugal, 1489-97.
19
Garson, G.D. (1991) Interpreting neural-network connection weights, A I Expert, 6(4), 47-51.
Gates, M. and Scarpa, A. (1978) Concreting and formwork logistics and optimisation, ASCE Journal of
the Construction Division, 104(CO2), June, 219-40.
Goh, B.H. (1996) Residential construction demand forecasting using economic indicators: a
comparative study of artificial neural networks and multiple regression", Construction Management and
Economics, 14(1), 25-34.
Goh, B.H. (1998) Forecasting residential construction demand in Singapore: a comparative study of the
accuracy of time series, regression and artificial neural network techniques, Engineering, Construction
and Architectural Management, 5(3), 261-75.
Goh, B.H. (1999) An evaluation of the accuracy of the multiple regression approach in forecasting
sectoral construction demand in Singapore, Construction Management and Economics, 17(2), 231-41.
Goh, B.H. (2000) Evaluating the performance of combining neural networks and genetic algorithms to
forecast construction demand: the case of the Singapore residential sector, Construction Management
and Economics, 18(1), 209-17.
Goldberg, D.E. (1989) Genetic algorithms in search, optimisation and machine learning, Addison-
Wesley, Reading, MA.
Haidar, A., Naoum, S., Howes, R. and Tah, J. (1999) Genetic algorithms application and testing for
equipment selection, Construction Engineering and Management, 125(1), 32-38.
Hegazy, T. and Ayed, A. (1998) Neural network model for parametric cost estimation of highway
projects, Construction Engineering and Management, 124(3), 210-18.
Hegazy, T. and Moselhi, O. (1994) Analogy-based solution to mark-up estimation problem, Computing
in Civil Engineering, 8(1), 72-87.
Kaka, A.P. and Price, A.D.F. (1991) Net cashflow models: are they reliable?, Construction
Management and Economics, 9(3), 291-308.
Kaka, A.P. and Price, A.D.F. (1993) Modelling standard cost commitment curves for contractors’ cash
flow forecasting, Construction Management and Economics, 11(4), 271-83.
Karni, R. (1987) A systematic approach to stochastic scheduling, Construction Management and
Economics, 5(3), 187-98.
Karshenas, S. and Feng, X. (1992) Application of neural networks in earthmoving equipment
production estimating, in Proceedings of 8th Conference in Computing in Civil Engineering and
Geographic Information Systems, ASCE, New York, N.Y., 841-47.
Karshenas, S. and Haber, D. (1990) Economic optimisation of construction project scheduling,
Construction Management and Economics, 8(2), 135-46.
Kenley, R. and Wilson, O.D. (1989) A construction project net cash flow model, Construction
Management and Economics, 7(1), 3-18.
20
Killingsworth Jr., R.A. (1990) Preliminary investigation into formulating demand forecasting model for
industrial construction, Cost Engineering, 32(8), 11-15.
Khosrowshahi, F. (1991) Simulation of expenditure patterns of construction projects, Construction
Management and Economics, 9(2), 113-32.
Levitt, R.E. (1987) Expert systems in construction: state of the art, Expert systems for civil engineers:
technology and applications, M.L. Maher, ed., ASCE, New York, N.Y., 85-112.
Li, H. (1996) Neural network models for intelligent support of mark-up estimation, Engineering,
Construction and Architectural Management, 3(1), 69-81.
Li, H. and Love, P. (1997) Using improved genetic algorithms to facilitate time-cost optimisation,
Construction Engineering and Management, 123(3), 233-37.
Li, H., Shen, L.Y. and Love, P.E.D. (1999) ANN-based mark-up estimation system with self-
explanatory capacities, Construction Engineering and Management, 125(3), 185-89.
McCaffer, R. (1975) Some examples of the use of regression analysis as an estimating tool, Quantity
Surveyor, December.
McCaffer, R., McCaffrey, M.J. and Thorpe, A. (1983) The disparity between construction cost and
tender price movements, Construction Papers, 2(2), 17-27.
McKim, R.A. (1993) A Neural Network Application to Construction Management, in Proceedings of
the Conference on Organisation and Management of Construction - The Way Forward, CIB W-65,
Trinidad, West Indies, September, 103-27.
Moselhi, O., Hegazy, T. and Fazio, P. (1991) Neural networks as tools in construction, Construction
Engineering and Management, 117(4), 606-25.
Moselhi, O., Hegazy, T. and Fazio, P. (1993) DBID: Analogy-based DSS for bidding in construction,
Construction Engineering and Management, 119(3), 466-79.
Newbold, P. and Bos, T. (1990) Introductory Business Forecasting, South Western Publishing,
Cincinnati.
Oshobajo, A.O. and Fellows, R.F. (1989) Investigation of leading indicators for the prediction of UK
contractors’ workloads (total new orders), Research paper, Construction Study Unit, University of Bath.
Pao, Y.H. (1989) Adaptive pattern recognition and neural networks, Addison-Wesley Publishing Co.,
Reading, Mass.
Paulson Jr., B.C. (1978) Interactive graphics for simulating construction operations, ASCE Journal of
the Construction Division, 104(CO1), March, 69-76.
Paulson Jr., Chan, W.T. and Koo, C.C. (1987) Construction operations simulation by microcomputer,
Construction Engineering and Management, 113(2), 302-14.
Pilcher, R. and Flood, I. (1984) The use of simulation models in construction, in Proceedings of the
Institution of Civil Engineers, Part 1, 76, Paper 8779: Engineering Management Group, 635-52.
21
Portas, J. and AbouRizk, S. (1997) Neural network model for estimating construction productivity,
Construction Engineering and Management, 123(4), 399-410.
Runeson, K.G. (1988) Methodology and method for price-level forecasting in the building industry,
Construction Management and Economics, 6(1), 49-55.
Sonmez, R. and Rowings, J.E. (1998) Construction labor productivity modeling with neural networks,
Construction Engineering and Management, 124(6), 498-504.
Tang, J.C.S., Karasudhi, P. and Tachopiyagoon, P. (1990) Thai construction industry: demand and
projection, Construction Management and Economics, 8(3), 249-57.
Tavakoli, A. (1985) Productivity analysis of construction operations, Construction Engineering and
Management, 111(1), 31-39.
Taylor, R.E. and Bowen, P.A. (1987) Building price-level forecasting: an examination of techniques
and applications, Construction Management and Economics, 5(1), 21-44.
Thomas, H.R., Mathews, C.T. and Ward, J.G. (1986) Learning curve models of construction
productivity, Construction Engineering and Management, 112(2), 245-58.
Thomas, H.R. and Yiakoamis, I. (1987) Factor model of construction productivity, Construction
Engineering and Management, 113(4), 623-39.
Touran, A. and Taher, K.A.H. (1988) Optimum fleet size determination by queuing and simulation,
Construction Management and Economics, 6(4), 295-306.
Tucker, S.N. (1986) Formulating construction cash flow curves using a reliability theory analogy,
Construction Management and Economics, 4(3), 179-88.
Wang, C.H. and Mei, Y.H. (1998) Model for forecasting construction cost indices in Taiwan,
Construction Management and Economics, 16(2), 147-57.
Wassermann, P.D. (1989) Neural computing: theory and practice, Van Nostrand Reinhold, New York,
N.Y.
Wijesundera, D.A. and Harris, F.C. (1989) The selection of materials handling methods in construction
by simulation, Construction Management and Economics, 7(2), 95-102.
William, T.P. (1994) Predicting changes in construction cost indexes using neural networks,
Construction Engineering and Management, 120(2), 306-20.
Woods, D.G. and Harris, F.C. (1980) Truck allocation model for concrete distribution, ASCE Journal of
the Construction Division, 106(CO2), June, 131-39.
Woolery, J.C. and Crandall, K.C. (1983) Stochastic network model for planning scheduling,
Construction Engineering and Management, 109(3), 342-54.