A Proposal Model for Estimation of Project Success in Terms of Radial Based Neural Networks: A Case Study in Iran

For improving the conditions of project intended purpose and reaching high score in the project success, project Stakeholders (including employer, contractor, consultant and its users) try to comply with the implementation of project Critical Success Factors(CSFs) at the beginning of each project. This implementation is in two terms: economic and executive. Artificial neural networks are one of the new methods which have been developed to estimate and predict parameters by using inherent relationship among data. In this research, it tried to propose a model to determine the project success score by using radial basis neural networks. For reaching this purpose, firstly, the key indicators of project success (employer, contractor and consultant) among the main elements involved in the industry of macro-civil construction projects in Iran reviewed. Secondly, ten CSFs key project success indicators were recognized in five categories: (i) financial, (ii) interaction processes, (iii) manpower, (iv) contract settings and (v) characteristic nature of the project (based on conditions of the present research in Iran). Then, some projects were selected by random sampling of projects operated during the last 5 years in the country's Ministry of Energy. Among those projects, project information was collected by managers of large projects. After training the designed neural network, the project success model was provided based on an assessment of project objectives including factors of Scope, Time, Cost, and Quality of the projects. For facilitating other researches’ use, the applied equation of the model was presented as well. Outputs, calculated by the proposed model, were in good agreement with the actual number of projects assessed in Iran. The results of this study may be used as a tool in implementing projects for the rapid assessment of achieving project goals’ facilities.


Introduction
The success of a project is of the largest and most important objectives and concerns of managers and all those involved in a project which is somehow unifying the efforts of all team members of the project.Reviewing the success and failure factors of projects in construction projects is more sensitive due to the dynamic and changing nature of the construction industry in various stages of project implementation.But, determining the success factors of a project is a complex and relative concept which most of the experts have proposed different and sometimes contradictory definitions due to their nature and execution system and natural characteristics.On the other hand, the volume of construction activities is of the essential development factors of a country.Annually, trillions of dollars of investments in the public and private sectors of various countries, either directly or indirectly, are expended in civil and building infrastructure.A construction project is a combination of different events, planned or unplanned, during the life cycle of the project, and survives under the umbrella of the changes in their environment.Among them, there are factors that are more important to the success or failure of a project.These factors are known as project success factors.In a definition, success factors of the project are expressed as "a set of environmental factors, facts or influential factors that can affect the output of the projects.These are factors that can accelerate a project or make it difficult.They can lead to the success or failure of the project but cannot be the basis for project evaluation [1,2].
Based on this, Project Management Body of Knowledge (PMBOK) standard is a unique effort to deliver a series of products (output) in the defined Scope, Time, Cost, and Quality [3].Projects include activities which should be carried out in the transparent scope and description of services, determined dates, with identified costs and determined quality; avoiding any of the four factors mentioned limits can lead to an unsuccessful and uneconomical project.
Defining a strategic framework that tactically examines the success of the project and identification of Critical Success Factors (CSFs) could be an important start point.As a result, the objectives of this assessment (1) are essential to identify the critical factors which overall determine project success (2) define and identify key CSFs of construction projects from the perspectives of different participants of the project with different goals.According to the actors in civil and forming CSFs for the project purposes such as scope, budget, schedule and quality, we can gain a better understanding of the project success.Management can take necessary steps to (1) avoid project failure (2) identify promising projects and keep track of them, and (3) identify the problematic areas of the project for undergoing necessary corrective actions

Definitions of Project Success
Researching on the project success has attracted the interest of many researchers and specialists.Tuman (1986) defined the success of the project as having anything that can be a reason for hope, predicting all project requirements and having sufficient resources to meet the requirements in the appropriate mode [4].In the same year, another definition was mentioned by De Wit.In the proposed definition, if a high level of satisfaction exists in terms of output and product of the project among key stakeholders including the main organizations, the project team and end users, it introduced as the overall success of the project [5].
In the following definition of project success, Ashley et al. (1987) suggest success as a consequence of better results, expected of the project (or normal level of project results), which is usually observed based on cost, schedule, quality, safety and satisfaction of participants.Since that time, safety is expressed a success factor [7]. Pocock et al. (1996), explained legal claims of the project as another indicator of the success of the project [8].Wuellner (1990) claimed that a project is successful when it is performed on time and on budget with a reasonable profit margin [8].
In addition, it should meet expects of customers and produce a high-quality project or consulting services while it limited the professional commitments of the company to an acceptable level.Summary of the definitions (provided by different investigators about the success of the project) is presented in Table 1.Sufficient focus on time, cost, and quality since such a definition entails a measurement of project success as too objective, difficult, and ambiguous due to disparity between project success and product success [11].

Critical Success Factors (CSFs)
The term CSFs, originally has been introduced by Rockart (1982) to define the number of activities that favourable results of which are necessary to achieve the objectives of project management [12].In other words, CSFs are factors which help to predict the success of the project in addition to its durability [13,14].A new definition was explained by Ogunlana and Toor (2009).It said CSFs means a specified element that helps considerably to the success of the project and is a very important component for success.Hence, for checking and ensuring the success of the project, one must be able to identify the factors that affect the success and failure of the project first and foremost [15].However, there is no general definition for CSFs or its evaluation.
Regardless of theory and empirical studies, success factors may also be identified by examining the actual projects.By taking the neural network approach, Chua et al. (1997) identified CSFs for the performance of construction budget [16].They were limited to quantifiable aspects and details of completed projects in the United States.Also, Chua et al. (1999) identified 67 elements related to the success from a professional survey focused on the budget, schedule and qualitative objectives of the project [17].It is reasonable to assume that the overall ranking of CSFs based on the involvement of participants in various projects will be different; hence it is considered as a study hypothesis.By forming CSFs for the purposes of the project (scope, budget, schedule and quality), the main players in construction projects can achieve a better understanding of the project success.On the other hand, the lack of sufficient and inclusive knowledge of project success factors makes controlling, monitoring, and performance of the projects difficult.Therefore, identifying effective factors which are important for the result of projects (depending on the type of projects) could provide a suitable framework for the evaluation of project outputs by managers and bosses and executives.Identification of factors in the success of the project can help to manage the appropriate allocation of resources over the life of the project as well [18].
According to the traditional projects, the potentially conflicting interests among stakeholders can also influence the success of an eco-city.The government and International cooperation are also important participants in an eco-city, which can facilitate the exchange of information and technical expertise on environment policy and innovation [19,20].Moreover, for identifying CSFs of eco-city and analyze the inter-relationship among them, Liu et al. opined that 'Clear definition and targets', 'conflicting interests among stakeholders', 'residents' behavior and awareness of environment protection and energy conservation' and 'management in operation stage of eco-city' were the most important CSFs for eco-city development in China [21].
Many developing countries face challenges to establish CSFs for public housing projects.Seven CSFs for PHPs in Nigeria is explained by Mukhtar et al. (2016) [22].These factors were: (1) Institutional framework for public housing, Also, the promotion of Green Buildings (GB) became a global trend and the CSFs of GB investigated.Shen et al. (2017) presented the importance of 16 CSFs of green building industry in Thailand [23].Using factor analysis, these CSFs of green building can be grouped into five categories, namely competence of project participants, integration of GB project team, technical and management innovation, external environment, and project characteristics.The results indicated that: 1) enhancing both competence of individual participants and integration of a project team are critical in fulfilling tasks of green building; 2) favourable global and local economic conditions can create new demand and supply of green buildings, and economic environment can greatly impact on the green building industry.Yang et al. (2017) created a list of CSFs for China's build-operate-transfer (BOT) projects to help the government and project companies to achieve success of BOT projects [24].This assessment shows that ten top factors in ranking analysis could play a useful role in the phases of preparing, bidding, constructing, operating and transferring BOT projects, which may be helpful for project companies and the government when making decisions.

Consolidated Framework of CSFs for Construction Projects
Pinto and Mantel (1990) conducted his research on the causes of project failures for 97 projects identified as failed projects [25].Sanvido et al. (1992) conducted a research to determine CSFs for construction projects by using data from interviewing with employers, architects, engineers, contractors [13].Mohsini and Davidson (1992) investigated the effects of violations stimulating organizational variables on cost, time and quality of the project and identified the most important determinants among these variables [26].In another study, Ahmed and Kangari (1995) reviewed factors which were understood as the most important factors by employers in contracting organizations [28].By studding on 280 construction projects, the relationship between the project performance and alternative approaches for managing the contractor-employer relationship was studied by Larson (1995) [28].In the case of contract investigation, By Alarcon and Ashley (1996), a method was proposed to model project performance [29].Prolongation of construction time in Hong Kong investigated by Chan and Kumaraswamy (1997) [30].
A research has been conducted by Iyer and Jha (2005) to identify CSFs affecting the cost performance of the construction projects in India [31].They reported coordination among project participants as the most important success factor for cost performance.Lam et al. (2008) investigated CSFs for construction projects with the method of construction and design [32].They identified nature of projects, effective project management practices and adoption of new management approaches as critical success factors CSFs for design and construction projects.Based on the results of another study by Pakseresht and Asgari (2012), identification and ranking of the important factors of success in construction projects of Pars Garma Company was carried out using a questionnaire survey among the 58 members of the management staff, project managers and technical experts [33].In the same year, with the aim of identifying critical success factors in public housing projects in Ghana and using the experiences of 13 specialists in the field of these projects, a study was conducted by Adinyira et al. [34].
In another study by Kog and Loh (2012), identification of CSFs conducted from the perspective of different specialists namely, civil and construction engineers, mechanical and electrical engineering, architects and monitoring devices [35].After collecting expertise comments from 27 building industry specialists, based on the Analytic Hierarchy Process (AHP), 10 superior and important factors for project success of CSFs were identified.Hong and Lim (2013) discussed about identifying (CSFs) of construction projects in Singapore.They prepared questionnaire about 32 important factors in the success of the project and distributed them among 12 specialists with at least 10 years of experience as representatives of employers, contractors and consultants [36].Factors affecting the success of construction projects in Malaysia (among 48 experts from contractors, consultants and developer companies) were examined by Yong and Mustafa (2013).They identified issues of the project manpower, commitment and communication of executive elements and management and control of project administrative process, were identified as the main topics contributing to success [37].

Artificial Neural Networks
Since of the processing way of the artificial neural networks is similar to the human brain functions and can have performances somewhat similar to the biological networks in learning, and extend this learning, the scope of research in this area continues to expand.The main idea of artificial neural networks is based on simulation of human's neural network.Neuron is the most important element of the processing in biological systems.There are more than 10 billion neurons in the human brain, each of which is linked to about other 10 thousand neurons and has created a dense structure.Each neuron contains input branches called dendrite, the main core and an output branch called axon.Axons of neurons are connected to the dendrites of other neurons at the synapse.When neurons become active, create electrochemical signals within its axon; this signal reaches other neurons via synapses and stimulates them.Each neuron becomes active when the total signals received by its body is exceeded a certain threshold.Figure (1) shows the overall structure and components of a neural neuron.According to various combinations of neurons and different learning rules, different networks are formed, one of the most useful of which in engineering sciences for nonlinear mapping, is the Radial Basis Neural Network (RBF) [38].

Structure Radial Basis Neural Network
Radial basis neural network was first used by Broomhead and Lowe (1998) as the stimulation function of neurons in the design of neural networks [39].The next researchers demonstrated that RBF networks are very powerful approximation mechanisms.By having a sufficient number of neurons in the hidden layer; RBF networks are capable of approximating any continuous function with any degree of accuracy requirements.In the following figure, the general structure of RBF network is shown.

Figure 1. The overall structure of a RBF network
In Figure 1, each I i represents one component of the input vector.Each σ i represents the area in which the i th neuron offers an acceptable and meaningful response, and u i is a vector the elements of which are weights connecting each of the inputs with the i th neuron.() is the excitation function of hidden layer neurons of an exponential Gaussian type as well.This excitation function is shown with equation 1. O j is also network outputs which are produced by applying a linear excitation function to the signals sent from the hidden layer to the output layer.
In equation 2, symbol T denotes vector transpose, and I is the input column vector.Each of communications shown in Figure 1 is established by a coefficient called weight.These weights are adjusted in the training process according to proper education rule.The number of hidden layers neurons is equal to the number of training pairs available in the training set.Also, the number of neurons in the output layer is equal to the number of components of the target vector.It should be explained that the training monitored using training pairs including input vector and the vector corresponding to the desired output, is conducted in a way that the difference between the actual outputs of the network and desired or objective outputs are minimized.In RBF network, the input layer weights do not require to be trained, and input matrix transpose is assigned as the first layer weight matrix.Weights of the second layer of RBF network are adjusted by the following supervised education act: 1.The input vector I i of the training set is delivered to the network.2. The Outputs of the hidden layer neurons are calculated and form the vector ().3. The network output vector O  is calculated and compared with the target vector t  .The second layer weight matrix w  is adjusted in order to reduce the difference of vectors O  and t  .To do this, the following formula is often used: In the above equation, η is the learning rate coefficient and is usually much less than 1. 4. Steps 1 to 3 are repeated for each vector of the training set. 5. Steps 1 to 4 should be continued until the error reaches an acceptable level and at this point, the training ends.After training the network by providing it with input vectors, the output vectors simulated by the network can be received [34].
To stop the computations in a neural network, two criteria are considered: (i)the number of performed periods be more than a limit, and (ii) the amount of error calculated in the output, be lower than its limit value.In this study, to evaluate the results of the neural artificial network, the criteria of mean square error MSE and factor of the goodness of fit (R, the correlation coefficient) were used [35].
In the above relations,   is the actual value of the variable,   is the predicted value of the variable,  ̅is the average value of the variable, n is the number of patterns, and  =   −  ̅  ,  =   −  ̅  .

Methodology
As mentioned in the previous section, factors associated with success used in the study, are mainly extracted from the results obtained by Chua et al. (1999) [17].This may be explained as, by removing the potential discrepancies in terms of social, cultural, political or other differences which may cause abnormal results, these results have been used in construction projects in Iran.In this study, it has been focused only on the main actors of the project namely the employers, contractors and consultants rather than covering all project participants.Due to relatively lower influence of subcontractors, suppliers and manufacturers on the success of the project on all construction projects in Iran, these actors have been removed.To collect comments data, a semi-structured questionnaire was prepared interviewing with experts and scholars in the construction industry of development projects in Iran, by choosing 41 indicators affecting the success of the project, in 5 components as Table 2. Due to the sensitivity and importance of the study, the questionnaire was distributed among experts and prominent executives and government officials and investors of macro projects which have the sufficient experience, expertise and knowledge in the management of construction projects.A total of 175 questionnaires were collected and after analysing the data and eliminating invalid questionnaires, 137 questionnaires were finally studied and analyzed.The effective factor of success of the project was evaluated based on Friedman test.The results of this line of research were calculated to examine the efficacy of the top 10 indicators among the 41 identified indicators in this study and were compared with previous studies in different countries.Critical factors of success in the country of Iran along with comparison of CSFs identified in this study with previous studies are shown in Table 3.Based on the results of this section, factors of "Insertion of realistic commitments and description of services and purposes specified in the contract" and "professional competence of project manager client" are as factors selected in the previous researches among experts in different countries [6, 17, 35, and 30].Also, due to much agreement between nature of Iran and Malaysia in terms of developing situation and shared spaces in construction projects in these countries, five indicators among the 10 key factors of the success of the project are in compliance.This compliance could be due to the developing situation in the two countries and sharing in the space of construction projects between the two countries.After identifying the 10 critical factors of success of the project, with the aim of finding a predictive model of success of the project in Iran using the radial basis neural network, we proceeded to design a new questionnaire to gather the actual information on the success of the finished project in Iran.In this questionnaire, the information was designed in two parts, Part I: model input based on prediction of achieving percentage of 10 critical factors of success of the project in five categories of financial, human resources, interactive processes, features of nature, and the contract items.Part II: the purpose of the model based on evaluation of the realization of the Scope, Time, Cost, and Quality of the project with preliminary estimates at the beginning of the project based on definition of the success of the project in the standard PMBOK [3].Since the accuracy of a model depends on the input data accuracy for network training, information needed to build the model and information of the construction projects of Ministry of Energy of Iran (completed over 5 years) were received from the ministry are shown in Table 4. Data of 80 projects were evaluated, and after removing the incomplete information, information of 56 projects was selected to build the model.

Table 4. A sample of the questionnaire
Project Target: Please complete the following information based on your project information.

Target
Questions Answer

Scope
Was the project completed in accordance with the scope and description of the services (including the initial Scope of the contract and the changes approved by the employer contained in the additions)?Yes□ No□ If the answer to the above question is No, Pls mention amount of deviation of completed project scope with initial contract scope?
Under 25% □ between 25% and 50% □ between 50% and 75% □ above 75%□ In this study, to simulate the structure of the artificial neural network and relating coding, the software MATLAB was used.In order to reach the success of the project using RBF neural network, the information of 56 development projects collected in the country of Iran, have been used for training and testing the network.Based on the information collected and the above description, 5 inputs of neural networks including scoring financial items, human resources, interactive processes, features of nature, and the contract items of each project were considered.The achievement level of each project objectives including scope, time, cost, and quality of the project with the estimation conducted at the time of the project start, were considered as 4 outputs of the neural network.The flow chart shown the estimation process of project success is presented in Figure 2.
Since the input values of the neural network have been scattered, and the possibility of lack of achieving the desired model due to scattering between the input data was expected, we limited input values in the range between zero and one through normalization by a division of all components of the input by their maximum.In the hidden layers of networks, a different number of neurons has been used and their optimal value has been determined to minimize error.Under the programming conducted, the number of hidden layers is begun with 1 and adding additional hidden layers continues until the increase in hidden layers does not affect the improvement of the error rate, which was selected as of 0.0159.

Description of Results
With the aim to build a prediction model for the success of the project in Iran, by radial basis neural network, it was proceeded to conduct the coding and analysis of data collected.In the program written for the production of neural networks, 4 outputs of scope, time, cost, and quality of the project were calculated with 4 distinct networks based on 5 parameters.The results of implementing the program on 56 projects in Iran, including correlation coefficient gained by comparing the calculated outputs with actual value outputs and the error rate calculated by Mean Square Error (MSE), are given in Table 4.In Figure 3, the number of Epochs calculated to achieve errors of trained data under the amount set as 0.0159, is shown and in Figure 4, the correlation coefficient obtained (Total Regression) of the trained network, and also histogram of network errors is shown in Figure 5.According to the results in the previous section, the outputs calculated by the proposed radial basis neural network show a good agreement with actual values of projects reviewed.Because of the complex network computations for evaluating basic data of each project, the neural network has been used to produce and provide an equation for determining the rate of success, based on calculation pattern.These complex network computations are the financial information, human resources, interactive processes, features of nature, and the contract items via weight and transfer functions within neural network and complexities involved in determining the success score of the project.This pattern is presented as follows:  = ‖ 1 − ‖ (6) y 1 = e −(n * b 1 ) 2 (7) PSI = w 2 * y 1 + b 2 (8) Given that in the success prediction model, radial neural network conducts the calculations in two layers.So the equation has 2 weighting matrices and 2 bias matrices of layers 1 and 2. In the above equation,  1 is the first layer weight matrix,  is the matrix of input data including 5 indicators of financial, human resources, interactive processes, features of nature, and the project contract items,  1 is the first layer bias matrix,  2 is second layer weight matrix,  2 is bias matrix of the second layer, (  ) is the score matrix of the success of the project calculated based on 4 output factors including the scope, time, cost, and quality of the project, and ‖ 1 − ‖ means calculation of the distance between the weights and inputs from the following relationship:

Table 4. Regression results and the calculated error
(A) Weight of the first layer including a matrix of 32 rows and 5 columns, are given in Table 5.
(B) Weight of the second layer including a matrix of 4 rows and 32 columns, are given in Table 6.
(C) Bias value of the first layer including a matrix of 32 rows and 1 column; all amounts of rows are considered as 0.832555.
(D) Bias value of the second layer including a matrix of 4 rows and 1 column, are given in Table 7.

Using the Proposed Model
For testing the proposed model and implementing program written in MATLAB, three other projects completed by the Department of Energy were randomly selected and investigated after the construction of the network.The projects included the installation of Irankhah 63 kV substation connections, construction project of access roads to the control dam of irrigation and drainage network plan of Chamshir, and governance and maintenance of drinking water facilities of the city of Neyshabour.All three projects have been exploited in the last 5 years.Information on each project was collected from project manager employers of all three projects as completed questionnaire forms.In the time of completing the questionnaires, all 3 project managers presented their project information together with the success grade of implementation of any project without the awareness of the designed model for calculating the success score of the project.Similar to section 5, information of three projects is entered in this project.After entering the information in the designed neural network, the correlation coefficient obtained in the assessment of the proposed success model presented (Total Regression Test) was calculated at a rate of 0.8689, and the error rate calculated by the Mean Square Error (MSE), (Total Performance Test) as 0.01088.In Figure 6, the correlation coefficient obtained (Total Regression) of the testing network.

Conclusion
This study was conducted to present a model to predict the success score of the project.For achieving this goal, we used radial basis Neural Network.The proposed model has been designed and presented based on studies and researches conducted in two stages including identification of factors relating to the success of the project and gathering the data of the exploited projects In Iran.As in the first phase, studies conducted to identify the success factors of the project, despite the general agreement on (CSFs) of the project.But due to geographical conditions of each region and conditions governing the space of the construction projects, there are no single factors among different countries.In this regard, and according to the project implementation status of Iran, 10 critical success factors of the project were recognized and selected among the 41 factors in previous studies based on researches conducted and performing semistructured interviews with experts and top specialists in the construction industry of Iran, from the view of employers, consultants and contractors, in 5 categories of 1-financial, 2-interactive processes, 3-human resources 4-contract settlement, and 5-characteristics of the nature of the projects.In the second stage of the investigation, information on the development projects exploited by project managers was collected through the Department of Energy.The data collected consist of two parts: a) the realization scores of 10 critical success factors of the as the model input and b) the realization of the objectives of the project according to the PMBOK, including Time, Scope, Cost, and Quality of the project (collected by the designed questionnaires) [3].After reviewing the information and using the neural network, the model for determining the score of project success was designed.Then, this model was evaluated by 3 other projects' models.There was a good compliance of the output values (calculated by the model with actual values).In this regard, the applied equation for use by other researchers was presented.This research can create practical applications for project leaders, which can exploit the results as guidelines for the formation of CSFs according to the customer's order and specifications for building projects.They can also compare the success factors identified in this study with real success factors in projects conducted in the past.This can be the use of a healthier knowledge management process and can be used more as a baseline for the formation and implementation strategies to increase performance.With a focus on predictive modeling of project success score in this study, project teams have better opportunity to access to important improvements in the performance of their projects and thereby, guarantee a promising future.

Figure 2 .
Figure 2. The flow chart of showing the process of the estimation of project success

Figure 6 .
Figure 6.Obtained correlation coefficient (Total Regression) of network testing data

Table 3 . Comparison of CSFs in different studies
During the implementation of the project, how much was the percentage of fulfilment of each of the following table based on the prediction of the above target.