Literature Review on the Techniques Applied in Each Frameworks
A review of the existing literature was performed on journal articles, conference proceedings, technical reports and online forums related to the area of study. The goal was to understand what has been done in the field of study and identify gaps to be filled. This research is about creating a decision support framework for construction projects and a navigational support system as a generic engine applied to the construction industry. In this research, for creating a framework to estimate project duration, multiple statistical and analytical tools, and, for NSS, multivariate statistical tools and dynamic decision-making tools, are used. A description of the methods and their applicability are given in the following section.
Decision Support System and Models
Decision support systems (DSS) were introduced by Gorry and Morton (1971) into the information system literature as “a computerised system that supports managers’ decisions in semi-structured decision situations”. According to D. Power (2002), decision support systems are categorised into five different types ranging from communication-driven, data-driven, document-driven, knowledge-driven and model-driven decision support systems. Communication-driven systems emphasise network and communication technologies, such as collaboration systems to support group decision-making tasks (D. Power, 2002). Data-driven DSSs use and manipulate external and internal time series data, real time data and online analytical processing (OLAP) (Codd, Codd,& Salley, 1993) to help decision makers at the operational or strategic level (D. Power, 2002). Document-driven DSSs emphasise different types of documents, such as oral, written and video by integrating a variety of storage and processing technologies to provide document analysis and retrieval. Model-driven DSSs (MD-DSSs) (Alter, 1982) emphasize access to and manipulation of financial, statistical and/or simulation models. Complex techniques are being used to create model-driven DSSs, such as decision analysis, mathematical programming and simulation (D. J. Power & Sharda, 2007). The majority of model-driven DSSs aim to find the desired decision criteria. In some MD-DSSs, forecasting techniques, such as time series analysis, are applied to bring accuracy to managerial decisions (Bermúdez, Segura, & Vercher, 2006). We will explain the usage of time series models in DSS in the next sub section. Knowledge-driven DSSs use artificial intelligence and statistical inference technologies to extract knowledge from databases and recommend actions to managers (Holsapple, Whinston, Benamati, & Kearns, 1996). Due to the development of tools which are available to support decision making, the need for a decision support system is vital in a decision making process (D. J. Power, Sharda, & Burstein, 2015).
Multiple-Criteria Decision-Making is responsible for structuring and solving decision and planning problems containing multiple criteria (Zionts, 1979). According to Majumder (2015), there are six steps in the decision-making process:
– Identifying the goal of the decision-making process.
– Selecting factors.
– Selecting available alternatives.
– Selecting weighting methods to identify the importance, such as analytical hierarchy (T. Saaty, 2008), analytical network process (Thomas L Saaty, 2001), and fuzzy multicriteria decision-making process.
– Methods of aggregation, including product, average or a function.
– Decision making based on aggregation results.
The multiple-criteria decision-making methods are categorised into two groups: compensatory methods and outranking methods (Billings & Marcus, 1983; Vincke, 1992). Compensatory decision making is systematic decision making that takes into account the importance of different attributes in decision making such as a linear model, AHP, Fuzzy logic decision-making (Billings & Marcus, 1983; Majumder, 2015). A noncompensatory model, such as elimination by aspects (EBA) (Tversky, 1972), is a model of a decision-making technique when decision makers are faced with several options. First of all a single attribute will be identified as the most important attribute for the decision maker. When the option does not meet the criteria for an attribute then it will be eliminated from being the most important and different attributes are applied until the best option is left. There are a variety of techniques available in noncompensatory models such as log-linear regression, non-linear regression and ANOVA (Billings & Marcus, 1983).
Analytical Hierarchy Process and Analytical Network Process
The analytical hierarchy process (AHP) is a structured technique for organising and analysing complex decisions; it was developed by T.L. Saaty (1990). It is based on mathematical and psychological concepts and is widely used around the world in a variety of decision-making situations. According to T. Saaty (2008), the AHP provides a comprehensive and rational framework (in a hierarchy) for structuring a decision problem, which is able to quantify its elements, relates those elements to the overall goals, and helps decision makers to evaluate alternative solutions. After framing the hierarchy, the AHP helps decision makers to evaluate the impact of each element by comparing them one at a time. AHP has been applied in different branches of construction, such as [a] selection of final constructor (S.-O. Cheung, Lam, Leung, & Wan, 2001), [b] evaluation of advanced construction technology (Skibniewski & Chao, 1992), [c] selection of critical success factors in construction management (D. Chua et al., 1999), [d] cost estimation of construction projects (An, Kim, & Kang, 2007) and procurement risk management (Hong & Lee, 2013). AHP is systematically adopted to determine the relative importance of elements to the goals of projects through a pairwise comparison of the elements. Generally, AHP modelling first breaks down a complex problem in the form of simple hierarchies, and then performs a pairwise comparison of the decision elements by computing their relative weights (F. K. Cheung, Kuen, & Skitmore, 2002). Although AHP is widely used in different parts of construction projects for decision making, this method suffers some deficiencies in practice due to its underlying assumptions. For example, AHP assumes that there are unidirectional relationships between criteria and subcriteria across the hierarchy level of elements. Also, it does not consider the correlation of different elements within each cluster. Hence, it fails to consider the interrelationships in each cluster of a component in a decision-making process (Thomas. Saaty, 2006). This is the reason which motivated development of the ANP model in a generic form which is known as the Analytical Network Process to overcome AHP limitations. The ANP technique is a suitable tool for decision-making when there are interrelationships between different elements (Thomas L Saaty, 2001). ANP allows all decision elements within each cluster to be compared pairwise in relation to the overall goal. If there are interdependencies among components of each cluster then the components should be compared pairwise. In the ANP method, a super matrix should be formed to show the priorities of all clusters and components. After forming the super matrix, the next step is ranking and prioritizing the alternatives by summing up the values of each column in the normalized super matrix and selecting the alternative which has the highest overall priority. We have used this tool in the module of multi-criteria decision-making to prioritise weather variables which effect project performance and project activities in Paper I.
Time Series Analysis
Time Series analysis is a statistical tool used to forecast data by monitoring essential variables over time. This tool is suitable when data are collected over time too, such as stock prices, sales volume, interest rates, weather and quality measurements. In time series analysis, there are several models that use the raw data to make a final model for forecasting. Auto regressive (AR), moving average (MA), auto regressive moving average (ARMA), exponential smoothing (ES), and autoregressive integrated moving average (ARIMA) are some of the models used. In theory, ARIMA modelling is the most general class of models for forecasting a time series which can be standardised by transformations (such as differencing and lagging) (G. E. P. Box, Jenkins, Gregory, & Greta, 2016). ARIMA modelling can take into account the trends, seasonality, cycles, errors and non-stationary aspects of a data set whilst creating or designing forecasts. In the ARIMA (p, d, and q) model, p is the number of autoregressive terms, d is the number of non-seasonal differences, and q is the number of lagged forecast errors in the prediction equation. For more details, readers are directed to Peter J. Brockwell and Davis (2002). Due to the dynamic nature of construction projects, the information should be analysed dynamically. The main objective of time series is understanding the dynamic or time-dependent structure of the observations of a single series (univariate) or multivariate series. The attributes of the dynamic structure are expected to contribute to an accurate forecast of future observations and to the design of optimal control schemes where it is required, such as process quality assurance (Peña, Tiao, & Tsay, 2001). The details of a times series model which is used to create a model-driven decision support system is explained in Paper I.
Regression analysis is a statistical technique to explore the relationships between variables. Usually this investigation reveals the causal relationship between a dependent variable and a set of independent variables (Sykes, 1993). There are different types of regression models, such as simple linear regression models, non-linear or polynomial regression, stepwise or multiple regression, logistic regression (Freedman, 2009), ridge regression, lasso regression (Tibshirani, 1996), and elastic net regression (Hui. Zou & Hastie, 2005). Simple linear regressions deal with only two variables, a single dependent variable and an independent variable, to find a linear function to predict the dependent variable values based on historical values of the independent variable. In data-driven decision support systems, regression models are used to predict the value of dependent variables (Pardoe, 2012). Multiple regression analysis considers the relationship between a dependent (criterion) variable with several independent variables (Hair, Black, & Babin, 2010).
Multidimensionality and Dimension Reduction Techniques
Principal Component Analysis (PCA) and Sparse Principal Component Analysis
Principal component analysis (PCA) is a statistical technique to reduce the dimensionality of datasets with p variables and n observations (I. T. Jolliffe, 2002). The main goal of PCA is to find a sequence of orthogonal factors (uncorrelated factors) from linear combinations of all variables. One of the main problems of PCA is, however, it finds a fewer number of important factors, and each factor is a combination of all variables, not only important variables. It is very difficult for a decision maker to understand which variables are important for one factor. Hui Zou, Hastie, and Tibshirani (2006) introduced a new method called sparse principal component analysis to solve this problem (Hui Zou et al., 2006). The lasso (Tibshirani, 1996) is a technique to reduce the dimensionality of datasets by considering possible zero loading factors. However, there are some shortcomings with lasso as stated by Hui. Zou and Hastie (2005). The number of variables selected by lasso depends on the sample size or number of observations. In other words, if the number of variables is more than the number of observations, the maximum number of variables that can be selected would be same as the number of observations. Hui. Zou and Hastie (2005) generalised lasso to solve the problem and called it elastic net where lasso is a special case of elastic net when the L2 penalty is 0. SCotLASS is another technique to find sparse loadings of factors (Ian T. Jolliffe, Trendafilov, & Uddin, 2003). The high computational cost is one of the shortcomings of SCotLASS (Hui. Zou & Hastie, 2005). PCA can be written in regression type (Cadima & Jolliffe, 1995). As mentioned earlier, each principal component is a linear combination of all variables, hence the loadings can be estimated from a simple linear regression model.
1.1 Background Problem
1.2 Construction Project Behaviour
1.3 Criteria to Measure Construction Project Success
1.4 current Tools for Monitoring and Controlling Project Performance
1.5 Aims and Objectives
1.6 Navigational Paradigm
2. Research Methodology
2.1 Methodological Requirements
2.2 A Review of Multi-Methodological Framework
2.3 Strengths and Definiens of Mentioned Frameworks
2.4 Methodology Adopted in This Research
3. Literature Review on the Techniques Applied in Each Frameworks
3.1 Decision Support System and Models
3.2 Multiple-Criteria Decision-Making
3.3 Time Series Analysis
3.4 Regression Analysis
3.5 Multidimensionality and Dimension Reduction Techniques
3.6 Overview of Three Papers
4.3 Literature Review
4.4 Conceptual Decision Support Framework
4.5 Case Study
4.6 Validation of the proposed Model
5. Paper II
5.3 Navigational Support System, Generic Framework
5.4 Research Methodology
5.5 Conceptual Framework of Navigational Support System
5.6 Prototype development using a case from an interior design project
5.7 Evaluation of proposed framework
5.8 Discussion and Conclusions
6. Paper III
6.3 Project Performance Measurement Systems in Construction Projects
6.4 Conceptual Model of Navigational Support System
6.5 Framework of Navigational Support System for Construction Projects
7. Discussion and Conclusion
7.1 Summary of Research Results
7.2 Link Between Papers I, II and III
7.3 Contribution to Research
7.4 Contribution to Practice
8. Future Research Direction
GET THE COMPLETE PROJECT