Diagnosing the success of the construction projects during the initial phases

Article history: Received October 25, 2015 Received in revised format: December 12, 2015 Accepted February 14, 2016 Available online Februray 14 2016 As construction projects are becoming more deployed and more complicated at the same time, having an instrument for anticipation of success has become a primary requirement for every stakeholder. On this basis, several models have been introduced which implement different methods for anticipation of the entire goals or a series of goals of projects. In this research, at the first step, 16 criteria as instruments of anticipation of success and 33 factors as required instruments for obtaining success were extracted through library studies, semi-structured interviews and the Delphi method. At the next step, by having 169 questionnaires filled by senior managers of construction projects, the importance and priority of each of these 16 criteria and 33 factors for the initial phases of projects were determined according to Iran’s local conditions. Ultimately, through modeling of data by a propagation neural network including 35 hidden layers, the anticipator model for success of construction projects during their initial phases was developed with Performance and Regression. This model is able to anticipate the level of realization of projects’ success criteria according to the level of realization of success factors at the initial phase. Growing Science Ltd. All rights reserved. 6 © 201


Introduction
Since realization of organizational strategies depends on the success of projects, the issue of success is crucial for the entire stakeholders in every project.Therefore, having an instrument, which provides a vision of projects' future has always been a stakeholders' demand.An instrument with the ability of anticipating the future states of projects have several applications including monitoring of projects during their lifecycles and facilitation of decision making processes.As a result, existence of different stakeholders with various objectives in a project, not only is a challenging task in terms of providing a universal definition for entire stakeholders' succes, but also it has led to formation of criteria with different and sometimes paradoxical priorities in the view of stakeholders.This issue is also disturbing in terms of providing a general and precise model for anticipation of projects' success.On the other hand, roughly long lifecycle of construction projects has led to introduction of errors into the results provided by anticipator models.
The purpose of the present research is to provide a model to forecast the level of realization of success criteria according to the level of realization of success factors at the initial phase.

Anticipation of project success
In early 1900s, the word success was, for the first time, academically introduced in the contexts of industry (Atkinson, 1999;Wit, 1988).The initial definitions were simple and qualitative.However, as the range of projects was expanded, more complicated definitions were developed with more deployed dimensions including both qualitative and quantitative concepts (Lavagnon, 2009).At first, the subject of assessment of success of finished projects and later, the subject of anticipation of projects which had not yet been started became the main subjects of researches in this context.Previous researches regarding anticipation of the success of projects, are ready to be categorized in terms of different perspectives irrespective of their selected method of anticipation:  Researches who tried to provide a model for predicting a single criterion such as time, expense, etc.For instance, Hwang (2008) used a regression model to forecast expenses.Also in another research, Hwang (2011) tried to anticipate the expenses of projects through Time series. Researches who tried to provide a model for predicting several criteria through defining a variable dependent on several other independent variables.For instance, Lipke et al. (2009) anticipated the time and expenses of a project through application of EVM (earned value management) technic.Also, Kim and Reinschmidt (2010) provided a model to anticipate project success through the earned value management (EVM) method. Researches who tried to develop performance and efficiency through considering different concepts as the basis.For instance, Li et al. (2006) tried to anticipate projects' potentials for violation of time tables and approved budgets through defining a potential function for a project as well as application of Fuzzy logics.In another similar research, the same researchers developed a model for design and construction contracts through application of regression (Lam et al., 2008).In another research, Cheung et al. (2008) tried to anticipate the index of satisfactory settlement of disputes through application of Regression method.In addition, Cheng et al. (2006) developed a model for anticipation of project efficiency through neural networks.

Implementation phases of construction projects
A project's lifecycle is a series of generally consecutive and occasionally overlapping phases whose names and numbers are defined by managerial needs of organization or organizations engaged with the project, project's nature and the contexts of application of project.Different phases of a project include subdivisions within a project which seek more control for effective management of completion of the main components of the project.Basically, different phases of a project are completed in an orderly and consecutive manner, however they can be overlapping in some projects too.
 Initial phase: this phase is an approval for the entire project.This phase includes whether a project should or should not start, estimates what has to be provided by the project, defines the overall goals of the project, defines customers' and managers and other engaged parties' overall expectations, defines the project's scope, selects the primary members of the project team, etc.  Planning phase: this phase defines and corrects goals and implements the best alternatives for obtaining the goals which are expected from the project.This phase includes determination of required resources, refinement of project area, provides the list of activities, determines the best sequence, provides an efficient plan and budget for allocation of resources, obtains the required approvals, etc.  Execution phase: coordination of people and other resources for execution of the program.This phase includes leadership of the project team, holding meetings with team members, communicating other important parties, trouble shooting, supplication of the entire required resources, etc.  Control phase: ensuring that project's goals are realized through continuous monitoring and measurements in order to spot violations and to be able to take necessary corrective actions.This phase includes monitoring of performance and reporting violations, making corrective decisions, performance of necessary reprogramming, limited reformations, etc.  Concluding phase: official acceptance of the project and advancing the project to a standard conclusion.This phase includes documentation of results, reassessing the process of project performance, reviewing experiences, writing the final reports, etc.

Neural networks
Nowadays, new methods of anticipation have been developed under the title of artificial intelligence (AI).AI is referred to systems which are able to manifest reactions similar to genuine humane reactions including perception of complex situations, simulation of thinking processes and humane reasoning methods, ability of knowledge acquisition and reasoning for problem solving.
Among different AI methods, neural networks is the most applied method for forecasting events.Neural networks are computational models which are able to determine the relationships between a system's input and output through a network of connected nodes.In these models, the level of activity of each of these connections is set according to chronological information (learning process) and ultimately, the model is able to explore the related laws between input and outputs whether being nonlinear and complex.Basically, a neural network includes the components of layers and different weights.Also the network behavior depends on the relation between participants, while there are three neuron layers in each neural network: Input layer: raw information are received and fed into the network.
Hidden layers: performance of hidden layers is determined by the input and with respect to the weight of the relation between the input and hidden layers.The weight of the relation between input units and hidden layers determines the timing for a hidden layer to become active.
Output layer: the performance of the output layer depends on the performance of the hidden layer and the weight of the relation between hidden layers and the output layer.
In a neural network, every two layers are connected by weights or in fact, fittings.There are various types of fittings or connections in neural networks.
There are different types of artificial neural networks with respect to the direction of entrance of information and processing of those information are divided into different types including Feed-Forward Neural Networks; Recurrent Networks; Radial Basis Function Networks and Multilayer Perceptron Networks.

Data collection
With the aim of defining and collecting the entire required criteria for assessment of a totally successful project on one hand, and also recognition and collecting the required factors for obtaining complete success from the views of the entire stakeholders of a project on the other hand; exploratory studies were performed in three steps.
At the first step, with the aim of collecting the results of previous researches among different scientific references, library studies were performed and resultantly, a series of criteria and factors of success were collected.At the next step, by carrying out six semi-structured interviews with experts of research domain, and by the help of a number of pre-determined questions, 6 more criteria and 14 more factors were added to the research.At the final step, with the aim of purification of 23 criteria and 47 collected factors, through the application of Delphi method and by incorporating the suggestions of five experts holding a Ph.D. and with sufficient amounts of executive experiences, the number of criteria and factors were respectively reduced to 16 and 33 which were suitable for Iran's Local situations.For this purpose, a primary questionnaire was developed and respectively, filled by the aforementioned five experts.
For the purpose of obtaining the main research goal, which was to provide an anticipator model, factors and criteria were prioritized according to Iran's local situation.For this purpose, a questionnaire comprised of two main sections was developed and sent for experts of construction domain.In the first section, the importance of each of 16 criteria were questioned for full assessment of success of a project and in the second section, the importance level of each of the aforementioned 33 factors were questioned in terms of obtaining complete success at the initial phase.On this basis, respondents were asked to give their answers according to a scale of zero (non-important) to four (very important).
The population of the research included entire project managers and planning experts of construction companies in the city of Tehran, Iran.A number of 260 questionnaires were distributed among the population and among this number, 91 questionnaires were excluded as a result of not being collected back or not containing the required validity.Diagram number one manifests abundance according to executive organs as well as experience level of the population.In addition, the aforementioned 16 criteria and 33 factors are mentioned in tables 1 and 2.

Removing the outliers
For removal of possible errors, outliers were identified after complete data collection.For this purpose, box diagrams were used as indexes for identification of outliers.These diagrams are used for manifestation of location, distribution and skewness of data.Through determining the first quarter (Q1), third quarter (Q3) and average, the IQR index was calculated for each and every one of the 16 criteria and 33 factors.According to this, each criterion or factor whose value does not fit the range of Eq. ( 2) is considered as outlier.
Q1-1.5IQR<values of criteria and factors<Q3+1.5IQR. (2) On this basis, the data were inserted into the SPSS software and resultantly, outliers were identified.Among the existing 169 questionnaires, 4 or more than four outliers were identified in 34 of the questionnaires and the aforementioned questionnaires were deleted for the purpose of reduction of errors and increasing the strength of the test.

Data analysis
The Kolmogorov-Smirnoff test was run for determining the normality of data distribution.The resulted significance for entire factors and criteria was less than 0.05 and this manifests the significance of the test or in other words, data are not normally distributed.Therefore, there is not any possibility of using parametric tests.On the other hand, since research variables are qualitative, non-parametric tests were used for analysis of data.
In addition, for the entire 16 criteria and 33 factors, the value of Chi-Square test was calculated with significance of 0.000.Since this significance level is smaller than error value of α=0.05, then we know that there exists a significant difference between ranks of each criteria.Table1 indicates the rankings of criteria according to Friedman tests' average ranking.Initial results of data analyses indicate the prioritization of three criteria of the iron triangle of the project namely as time, costs and quality.These criteria have acquired the highest ranks among other criteria.In addition, three criteria of Minimal changes in organizational culture due to the project in the organization, All participants' consideration of project success and the least effectiveness of the project from environmental factors (political, economic and cultural) have acquired the lowest of ranks among other criteria.
In addition, a ranking similar to criteria is performed for required factors for obtaining success during the initial phase.The present ranking is based on the scores obtained from Friedman's test which is a non-parametric test.Primary analysis of factors indicates the priority of three factors including awareness about the project, clarity of project specifications and financial support reference over other factors in the first or initial phase of the project.On the other hand, factors of attraction of labor force, clarity and speed in providing technical solutions and having an experience of application of quality control services have acquired the lowest of all ranks among other factors.The noteworthy point here is that the performed ranking indicates the priority of criteria and factors according to Iran's local conditions and situations during the initial phase of projects.

Model presentation
The presented model in this research is comprised of two main sections.The first section is concerned with development of a model for determining the relation between the input and collected output through application of neural networks and the second section is concerned with transformation of the obtained results for determining the value of anticipation for realization level of success criteria from the level of realization of success factors.Fig. 2 indicates the main levels of model in addition to subdivisions of each level.

Development of the Model
As it was mentioned earlier, the collected data on one hand determine the importance of each factor for obtaining a hundred percent success during the initial phase of projects and on the other hand, determine the level of importance of each criterion from the perspective of every expert.On this basis, through application of neural networks, we can present a model for determining the relationships of the factors for the success and criteria of success.This model is the sum of views of experts regarding the relation between these elements.By using the neural Network toolbox in the software of MATLAB, a multilayer perceptron network was developed for the aforementioned model.For this purpose, the software was tested fifty times and the most optimized and precise model was selected.The selected model is a network with two hidden layers and respectively 50 and 16 neurons in each layer.The whole 33 inputs of this network indicate the importance of factors and its 16 outputs also indicate the importance of each criterion.The importance level of factors and criterions are reflected on a scale of 1 (non-important) to 5 (very important).Fig. 3 provides a view of the components of model.The number of inputs for the purpose of network testing and training was equal to the number of effective questionnaires as 134 series.Of this value, 70% is used for training, 10% is used for testing and 10% is used for validation of the model.Also the Bayesian Regularization algorithm was employed for development of this model.
Once the network was developed, the error level was calculated through Eq. ( 3).

Errors= targets-outputs
(3) Highest error distribution is between the values of -0.11 and 0.2.The error distribution diagram for the developed network is shown in figure 4.   The values of aforementioned indexes indicate the suitability and desirability of the precision of the proposed model.This means that this network is ready to be employed as a model for determining the importance of success factors according to the importance of success factors.

Transformation and interpretation of results
Since the purpose of this paper was to anticipate the level of realization of success criteria according to the level of realization of success factors, the next step is concerned with providing a method for implementation of the proposed model for obtaining this goal.
For realization of the aforementioned goal, this level includes four steps.The first step is concerned with reviewing and analyzing the existing conditions.For this purpose, 33 factors required for obtaining success should be analyzed and the level of realization of each of these factors should be determined according to Table 3.The next step is concerned with putting the level of realization of each factor into effect.Eq. ( 4) indicates the effect of level of realization of each factor on their consequent importance.
In Eq. ( 4), the matrix F indicates the level of each factor's importance, the Wi matrix indicates the level of realization for the i factor and the f matrix indicates the effect of conditions on importance of each factor.
The third step is concerned with employing the matrix on the extracted model as the input.The model's output in this step is a 16×134 matrix.
The final step of this level is to recover the outputs of the model to the forecasting value of criteria fulfilling.For this purpose, at the first it is required to divide each entry of the output matrix of neural networks by its corresponding entry on the matrix of importance of criteria.Eq. ( 5) indicates the manner of calculation of level of realization of criteria.
In this relation, the p matrix is the matrix of anticipation of criteria's realization, the c matrix is the network's output matrix and the C matrix is the matrix of importance of criteria.
At the end, prediction values of each success criteria are evaluated.For this, average elements of each line of p matrix that is obtained from relation 6, is as results of proposed model that is presented in this research.
∑ 134 , where Pi is the element of ith line that is forecasted and Pj; which is the jth criteria that is evaluated.
As is described before, P is the amount for each criterion, which shows how many of them are realized.A 0 value for P means no absolute amount and one for P means complete amount.So, when the amount of P for a criterion is equal to zero, forecasted result is weakness for the realization of that criterion.
Based on sensitivity analysis, on one hand, the first three criteria (C1, C2, and C3) cover 34% of the variance of project success, but in the other hand, first five factors (F1 until F5) cover 47% of the variance of essential factors to success, that these reflect the importance of these three criteria and five factors in projects.
Interpretation of P amount on the one hand, and satisfaction level for each criteria on the other hand are related to kind of users and Stakeholders of model.Accordingly, different Stakeholders with various strategies and with different kinds of priorities for all criteria can reach different results.
With P values that are calculated and importance of each criteria, judging of project success can be predicted.This subject can be as suggestion for future research,

Model validation
Model validation was conducted to ensure that the constructed model in this research is generalizable to the population and not specific to the sample used in training.
First, R, R 2 and Mean Square Error (MSE) for the network are calculated, While 10% of the initial data were used to assess neural network of model.Results are shows in Table 4.These results confirm high reliability of model.Also a diagnostic test was conducted and the residual plots of R (being actual C less the predicted C) versus predicted C showed a random distribution.This indicates that the normality assumption is valid.

Fig. 1 .
Fig. 1.Abundance according to executive organ and experience

Fig. 3 .
Fig. 3.A view of the model developed with neuron networks

Fig. 4 .
Fig. 4. Error distribution diagram In addition, the model's performance index is equal to 0.0180.This value was obtained in cycle 988 among 1000 training cycles.Fig. 5 indicates the changes of performance index during the network's training period.

Fig. 5 .
Fig. 5.The changes of performance index during the network's training period Another important index in terms of neural networks, is the index of regression and for the proposed model, this value is equal to 0.9267 for entire inputs.Fig. 6 presents the processed regression diagram.

Table 1
Acquired ranks by criteria

Table 2
Collected success factors

Table 3
Guide to determination of factors' realization level

Table 4
Performance Evaluation of model's neural network