Estimating the Optimum Duration of Road Projects Using Neural Network Model

The aim of this study to predictedthe duration of road projects in republic of Iraq. Historical data was adopted for (99) projects for interval between 2000 to 2017 from Roads and Bridges Directorate (RBD). Artificial Neural Network (ANN)model used to estimate the duration usingsix variables (length of road, No.of lane, No.of intersection, volume of earth, type of pavement and furniture level). The methodology used in this study included two important parts, the first part, reviewing the literature of the subject (estimating the duration of the road projects), and the second part, used of a program neuframe v.4 to build the models of neural networks to estimate the duration of road project. The results showed strong correlation between actual duration and predict duration by (90.6%), minimizes testing error (3.2%) and training error (4.9%). The MAPE and Average Accuracy Percentage generated by ANN model were found to be(25.73 %) and(74.27%) respectively. Therefore, it can be concluded that ANNs model show very good agreement with the actual measurements.


III. RESEARCH JUSTIFICATION
Research justification can be summarized follows: 1) The scarcity of studies and researches related to the field of artificial neural networks in estimating the length of road projects, which makes this study an original addition to the scientific knowledge in the field of project management, as the results of this study may be a strong incentive for conducting original studies in the future.
2) The lack of reliability of methods and techniques used in estimating the length of road projects in the construction sector in Iraq. As a result, there is an urgent need to use sophisticated modern methods and techniques for the purpose of estimating the duration in road projects based on computer models or mathematical equations of high accuracy IV. RESEARCH LIMITS The limits of this study were defined as follows:

VI. APPLICATION ANN IN ESTIMATION THE DURATION OF CONSTRUCTION PROJECT
Studies and research on the subject of neural networks and their relation to the management of construction projects in the construction sector in the Republic of Iraq are few. The researcher studied a number of these studies and researches, including the study of( [5], [6], [7], [8], [9], and [10]); the researcher finds that the vast majority of this research was concentrated in the subject of estimating of the construction productivity and cost estimating of the construction projects.
The process of estimating depends on personal experience or on historical data from previous projects, and the engineer who makes the estimation was not based on a mathematical equation with high accuracy in the calculation of construction productivity or in calculating the cost or duration or other items.
Iraqi studies and research did not address the issue of calculating the duration of road projects using smart artificial neural networks according to the researcher's knowledge, except for one attempt by the researcher [11], which was the subject of this study on the estimation of the duration of the implementation of the projects of concrete irrigation channels. As for the Arab and foreign studies and researches, the researcher has studied a number of them, especially those that are based on calculating the duration of the construction projects for the different items based on the artificial neural network technology as shown in Table I; those previous studies did not address the development of mathematical equations accurate. Therefore, this research is considered an attempt by the researcher to enter into this type of researches to calculate the duration of road projects with high accuracy based on artificial neural networks, and the importance of this research was considered a complementary series of previous research in the field of application of neural networks in the project management, but the advantage of this research lies in the use of the neural network (Perceptron) , in addition to the study of a number of more influential variables, and concludes this research to find a mathematical equation in order to calculated of the duration of road projects The construction project is investigated to demonstrate the use and capabilities of the proposed model to see how it allows users and experts to actively interact and, consequently, make use of their own experience and knowledge in the estimation process. The proposed model is also compared to the well-known intelligent model (i.e., BPNN) to illustrate its performance in the construction industry

VII. FACTORS AFFECTING THE ESTIMATION OF THE DURATION OF ROAD PROJECTS
The identification and evaluation of the factors that affect the estimation of duration is a critical issue facing the evaluators in order to increase productivity in road projects. Understanding the critical factors that affect the estimation of the duration, whether positive or negative, can contribute to the development of a strategy to reduce the shortcomings and improve the effectiveness of the performance of the construction project. There is an urgent need to identify and understand the various factors that affect the duration of road projects, in order to focus the necessary steps in an effort to reduce the cost of the project and delay the completion of the project, thus increasing the productivity and overall performance of the road project.
Historical data is collected from completed road projects in Iraq from 2000 to 2017. The researcher succeeded in gathering well trusted data for more than 99 projects through visited the Roads and Bridges Directorate (RBD and reading the concerned sheets, documents and reports for road projects. Table II,

VIII. DEVELOPMENT OF A MATHEMATICAL MODEL USING THE TECHNOLOGY OF NEURAL NETWORKS
In this study, the researcher used an integrated scientific methodology to build the neural network model to estimate the duration of road projects. This methodology includes the following steps: 1) Selection software There are much software used in the field of neural networks, including program (Matlab), (GMDH Shell) and (Neurosolutions), but the most common program for researchers is NEUFRAME V.4. This program is easy to use and is far from complicated. NEUFRAMEis an integrated group of Intelligence Technology tools that include Neural Networks logic that allow putting the power of neural nets to work straight out of the boxand the Fig. (1) below illustrates its architectural layout.

2) Input Actual Data
The process of selecting the variables in the input and output icon is of great importance that contributes to the improvement of the performance of the neural network. The increase in the number of input and output variables has a significant effect on increasing the size of the neural network, thus reducing the speed of the learning process and thus affecting the efficiency of the neural network.
There are several ways to select the number of variables in the input and output model and the Method of Priori Knowledge was chosen in this case, as this method is widely used in the construction sector and is approved in many researches and studies. This method can be used when there is no prior knowledge of the input variables and their effect on the output variables. Therefore, the input model included the following variables: F2: Length of the road,F3: Number of standard lanes,F4: Volume of earthworks,F5: Number of intersections,F6: Types of paving, F7: Road furnishing level. While the output model included one variablewas F1: Duration of road project. Fig. (2) shows inputs and outputs in the NEUFRAME program.

3) Data Division
Input or output data in the neural network are either continuous variables or discrete variables, and these are divided The data is divided into three main groups: 1) Training group to build the neural network model (Training Set.) 4) Testing group or testing the neural network model (Testing set.) 4) Validation group to estimate the performance of the model in the applicable environment.
The training group is used to adjust the weights connected to the neural network. Testing group group is used to check the performance of the network at different stages of education, and the training is stopped when the error of the examination group increases. The validation group is used to assess the performance of the model once the neuronal network training has been successfully completed. Therefore, dividing the data into the three groups above is a critical and important step in neural network modeling. In this research, the statistical consistency method was used for the purpose of dividing the data into the three groups (the training group, the Testing group and the Validation group). This method ensures statistical suitability of the data for each group, thus ensuring that there is no bias in dividing the data in each group using (T-test) through the use of statistical standards, namely the arithmetic mean and the standard deviation and range. The advantages of this method are that it adopts the trial and error method to reach the best division of data. Fig. (3) shows the percentage of data breakdown for the training, testing and validation groups using the trial and error method. The researcher used different percentages of data for these groups in an attempt to obtain the best performance of the neural network, namely, reaching the highest correlation coefficient to show the strength of the relationship between the output of the neural network and measured output and in conjunction with the lowest error rate for the testing group, these criteria are adopted in this research to choose the best division of data.  Fig. (3) shows that the best division of data is 73.74% for the training group, 6.06% for the testing group and 20.2% for the validation group, with the lowest error rate of the test (3.2%) and the largest correlation coefficient (90.6%), as shown in Fig. (4).

Fig. 4. Training Error and Testing Error
For the purpose of distributing the total data of the 99 variables, the three groups, namely the training group, the testing group and the validation group, the Neuframe program provides an efficient way to distribute the data in three ways: 1) Random mode: In this method, the program randomly distributes variables data on the three groups and according to the percentages obtained in Fig. (3) 2) Strip mode: In this method, the program divides the total data into non-specific sets of packets, and each package includes data for the training group, the testing group and the validation group, 4) Blocked mode: In this mode, the total data is treated as one packet and divided by the three groups. The first 73.4% of the data is for the training group and 6.06% for the testing group, and 20.2% for validation group as shown in Fig. (3) In order to study the effect of the use of different options (random, blocked, strip) as shown in Fig. (5), it can be observed that the best performance of the neural network is when using the division method blocked, where it has the least error for the test (3.2%). Using the default parameters for the program used in this research is the Learning Rate 0.2 and Momentum Term values 0.8 and the Transfer Function in the output layer and the hidden layer is Sigmoid, Thus, the typical form of this network developed in this research is a three-layer neuron is an input layer and includes six neurons and a hidden layer hidden layer comprising three hidden nodes and output layer with only one neuron as shown in Fig. (6) and Fig. (7). The small number of connection weights obtained by Neuframe for the optimal ANNs model enables the network to be translated into relatively simple formula. To demonstrate this, connection weights and threshold levels (bias) are summarized in Table II. X1= {θ7+ (w7-1*F2) + (w7-2*F3) + (w7-3*F4) + (w7-4*F5) + (w7-5*F6) + (w7-6*F7)}………………..…(2) X2= {θ8+ (w8-1*F2) + (w8-2*F3) + (w8-3*F4) + (w8-4*F5) + (w8-5*F6) + (w8-6*F7)}……….…… (3) X3= {θ9+ (w9-1*F2) + (w9-2*F3) + (w9-3*F4) + (w9-4*F5) + (w9-5*F6) + (w9-6*F7)}…… .….… (4) It should be noted that, before using Equation 2,3 and 4, all input variables need to be scaled between 0.0 and 1.0. It should also be noted that the predicted value of the duration obtained from Equation 1 is scaled between 0.0 and 1.0 and in order to obtain the actual value this duration has to be re-scaled. The procedure for scaling and substituting the values of the weights and threshold levels from Table II Table III, the MAPE and Average Accuracy Percentage generated by ANN model were found to be 25.73 % and 74.27% respectively. Therefore, it can be concluded that ANNs model show very good agreement with the actual measurements. To assess the validity of the ANNs model, the predicted values of Duration are plotted against the measured (observed) values of duration for validation data set, as shown in Fig. (8). It is clear from Fig. (9). The generalization capability of ANN model uses the validation data set. Coefficient of determination (R 2 ) equal to (82.2 %), therefore it can be concluded that ANNs model show a very good agreement with actual duration.

IX. CONCLUSION
The main aim of this study is to use a new approach known as artificial neural networks model to predicting the optimumduration of road projects in field of project management inRepublic of Iraq. The application of artificial neural networks as a new method in the project management was very necessary to ensure the success of project management. One model was built to predict the duration of the road. In this study, multilayered networks were used in the post-error propagation approach. It was found that these networks have an excellent predictability of 74.27% as average of accuracy and correlation coefficients (90.6%).