Construction Output Forecasting: A Comparison of Econometric and Artificial Intelligence Modelling Techniques

建造產出量預測模型: 計量經濟學和人工智能建模方法的比較

Student thesis: Doctoral Thesis

View graph of relations

Detail(s)

Awarding Institution
Supervisors/Advisors
Award date14 Aug 2017

Abstract

The construction industry contributes to the national economic growth process. Construction output is the comprehensive measure of the volume of investment in construction projects at industry-level. Construction output tends to fluctuate over time due to economic cycles. These fluctuations have an adverse effect on the construction industry and the economy. Reliable prediction of changes in the volume of construction output would facilitate development of strategic plans for the construction industry. Econometric techniques are the most popular approach used for modelling and forecasting of construction output. However, the limitations of these models include: inability to capture the non-linearity which is common in real-world problems, the need for complex statistical tests for model verification, and the difficulty in distinguishing between 'causality' and 'correlation'. In view of these problems, the present study was undertaken to examine the efficacy of using artificial intelligence techniques for modelling and forecasting of construction output. Construction output was disaggregated into seven parts: overall, private, public, residential, commercial, industrial and maintenance construction output. Broadly speaking, the models applied in the present study can be classified into two categories, namely: univariate and multivariate. Also, the techniques utilized are Box-Jenkins (BJ), vector error correction (VEC), artificial neural network (ANN) and support vector machine (SVM) model. The study was carried out in five distinct but interrelated phases: (1) literature review, (2) data collection and treatment, (3) variable selection, (4) model fitting and (5) comparison of forecast accuracy of the developed models. Out-of-sample forecasts generated by the models provide an objective measure for evaluating the reliability of the developed models. The results show that changes in the volume of construction output can be predicted by economic variables. In addition, artificial intelligence (ANN and SVM) models produce more reliable forecasts of construction output when compared with econometric (BJ and VEC) models. On the overall, the ANN models produce the reliable forecasts of construction output in the different segments of the construction market. This thesis contributes to the knowledge on construction output forecasting in several ways. First, the artificial intelligence techniques extend the methods available for construction output forecast. Second, the study provides insights on the factors that influence the volume of investment in the different segments of the construction market. The models can be used as a tool for estimating the future movements in the volume of construction output.

    Research areas

  • Artificial intelligence, Construction Output, Modelling, Econometrics, Forecasting