الفهرس 
Chapter (1) IntroductionOnly 14 pages are availabe for public view
 |  | from | 126 |  |  |
| | | from | 126 | | |
Abstract
One of the most important tasks confronting planners in the construction industry is the risk management. Project risk management includes: risk management planning, risk identification, qualitative risk analysis, quantitative risk analysis, risk response planning, and risk monitoring & control (PMI 2008). One of the means to mitigate the risks of construction projects is the contingency. If cost contingency is too high, it might cause the construction project to be uneconomic, aborted, and lock-up funds not available for other organizational activities; if too low, it may result in unsatisfactory performance outcomes. In this research, the author attempts to introduce a powerful and simplified models for predicting the cost contingency for steel reinforcement material in building projects based on Regression Analysis (RA) and Artificial Neural Network (ANN). Many researches have proposed different methods of cost contingency. The methods of contingency calculation are divided into three main groups:” Deterministic methods, Probabilistic methods and Modern mathematical methods” (Payam and Touran 2014). Deterministic methods include” Traditional Percentage and Factor Rating & Estimate Quality”. Whereas probabilistic methods include “Monte Carlo Simulation, Method of Moments, Individual Risks, Expected Value, Range Estimating, Influence Diagram, Regression, Theory of Constraints, Analytical Hierarchy Process and PERT ”. On the other hand modern mathematical methods include” Artificial Natural Networks, and Fuzzy Sets”.Monte Carlo Simulation (MCS) captures outcome of risk identification and impact which can be used to estimate contingency (Razali and Fabi 2017). RA is more accurate than traditional method. ANN is suitable for non-linear modelling of data, which contrasts with the linear approaches using regression (Chen and Hartman 2000). Thus, in the current research, the author intended to make use of both the benefits of MCS in its accuracy and RA in its simplicity and grouping all the factors that drive all risk factors to predict Simulated Cost Contingency (SCC). Also, to make use of both the benefits of MCS in its accuracy and ANN in its suitability for non-linear modeling of data to predict SCC. The objective of the current research is to develop ANN and regression based models to predict the simulated cost contingency due to the variability of steel reinforcement prices. The main contribution to the body of knowledge lies in the fact that the developed models’ predictions are expected to serve as a useful basis for contractors to make easier calculations of SCC for building projects due to variability of steel reinforcement prices. Thereby facilitates incorporating cost contingency into markup for building projects. Also, for researchers to adopt the same methodology for other types of projects and for assessing other materials prices variability simultaneously. Two scenarios are adopted in developing the models. In the first scenario (case 1), seven independent variables which are sensitivity analysis ratio, direct cost of steel reinforcement to total project direct cost ratio, two probability distributions (triangular and normal), and three types of contractor’s trend in dealing with risk (gambler, neutral, and conservative) were applied. In the second scenario (cases 2-7), two independent variables only which are sensitivity analysis ratio, direct cost of steel reinforcement to total project direct cost ratio for one type of probability distribution and one type of contractor’s trend at a time were used. Since, cases 2, 3, 4, 5, 6, and 7 correspond to (Triangular & Gambler, Triangular & Neutral, Triangular & conservative, Normal & Gambler, Normal & Neutral, and Normal & conservative), respectively. In ANN models development, best transfer and learning functions were determined. In scenario 1, among 75 trials which are performed to determine the best network architecture, a multilayer feed forward NN formed of two hidden layers, the first consists of six neurons where the second one contains two neurons is found to be the best NN architecture. In scenario 2 (cases 2-7), 12 trails are carried out for each case. In all the cases, a multilayer feed forward NN formed of one hidden layer contains 2, 2, 4, 6, 15, and 6 neurons corresponding to cases 2, 3,4,5,6, and 7 respectively is adopted. Two types of validation are performed for the best models. Validation type I for comparing the predicted simulated cost contingency with determined simulated cost contingency calculated by MCS. Validation type II for comparing the predicted simulated cost contingency with the real values of cost contingency. Analysis of the results show that: 1. Validation type I of ANN based model for comparing the predicted simulated cost contingency with determined simulated cost contingency revealed that the value of Mean Absolute Percentage Error (MAPE) for scenario 1 (Case 1) equals to 23.1%, whereas it has an average value of 12.1 % for scenario 2 (cases 2-7). 2. Validation type II of ANN based models for comparing the predicted simulated cost contingency with the real values of cost contingency revealed that the value of MAPE for scenario 1 (case 1) is 30 %; whereas it has an average value of 31.4 % for scenario 2 (cases 2-7). 3. The best model of the regression model in scenario 1 (case 1) included five independent variables which are sensitivity analysis ratio, direct cost of steel reinforcement to total project direct cost ratio, normal distribution, gambler contractor, and conservative contractor, whereas both triangular distribution and neutral contractor were excluded. In the second scenario, one model is extracted for each case among the cases 2-7. 4. Validation type I of the best regression model revealed that MAPE for scenario 1 equals to 26.5 %, whereas it has an average value of 22.4 % for the six cases included in scenario 2. 5. Validation type II of the best regression models revealed that MAPE for scenario 1 (case 1) is 24.7 %, whereas it has an average value of 21.5 % for five cases (2, 4, 5, 6, 7) among six cases of scenario 2. 6. ANN models are more powerful than regression-based models in theoretical modeling of cost contingency, whereas regression-based models are superior from practical point of view. The contractors are advised to use regression based models as these models outperform ANN models for the practicality in estimating cost contingency. Future works is recommended in the following aspects: (1) development of similar regression-based models for predicting the simulated cost contingency for different construction materials simultaneously and for other types of construction projects, (2) adopting other techniques such as Fuzzy Set Theory, and Random Forster in prediction of simulated cost contingency and comparing them with regression-based models developed in this research.