Abstract:
The objective of this research is to create models for predicting the construction cost indexes, published in Engineering News Record (ENR), in order to obtain the values that support more accurate preliminary estimation. The study focuses on the prediction of the changes of two construction cost indexes: Building Cost Index (BCI) and Construction Cost Index (CCI). The ranges of the prediction are one month and one year in advance. The methodologies used in this research consist of: (1) collecting construction cost indexes data and information relevant to the development of such indexes, (2) developing artificial neural network models for predicting the construction cost indexes including Back Propagation (BP) and Neuro-genetic Algorithms (NGA) neural networks, and (3) comparing the predicted results from artificial neural network (ANN) models with those from ENR and statistic methods such as a Simple Average, an Exponential Smoothing and a Simple Regression Analysis. Besides, this research also applies the development method of the proposed models to the develop models to predict the construction material price indexes in Thailand. This research shows that the ANN model that predicts the BCI one-month in advance provides results with RMSE of 16.365 while a prediction using a statistical model provides results with RMSE of 16.692. The ANN model for predicting the CCI one-month in advance has RMSE of 28.745 whereas a prediction using a statistical model has RMSE of 29.248. In addition, in the case of predicting indexes one-year in advance, the proposed ANN model provides results with 20.744 of RMSE for BCI prediction whereas a statistical model prediction yields RMSE of 68.061. The ANN model that predicts the CCI one-year in advance has RMSE of 39.288. On the other hand, a statistical model prediction's RMSE is 58.986. The comparison of ANN model predictions and the ENR's predictions indicates that the ANN model for predicting BCI one-year in advance has RMSE of 11.068 whereas the ENR's prediction has RMSE of 43.252. Additionally, the proposed ANN model for predicting CCI one-year in advance yields RMSE of 26.669 whereas the ENR's prediction yields RMSE of 50.374. When the validated data sets are applied to the Artificial Neutral network models, the ENR method and statistical, it shows that the prediction using the ANN models yields the most accurate results of all