Building transparent construction performance models via techniques of computational intelligence

Building transparent and highly interpretable models of the construction performance is generally of significant importance to construction managers. However, previous research focuses more on the approximation accuracy of construction performance models. Few studies have been done on the transparency of models, i.e., offering some understandable cause-effect relationships between the construction performance indicator and its influence factors.; The general objective of this thesis is to build transparent construction performance models using techniques of Computational Intelligence (CI). More specifically: (1) First, a neural network, named General Regression Neural Network (GRNN) is selected as the basic modeling technique. Its new genetic algorithm based learning algorithm is introduced. The GRNN not only presents a high approximation rate, but also offers importance indices about the influence of inputs on the output and (2) Secondly, a fuzzy clustering algorithm is introduced to granulate the inputs into their linguistic terms. The model built with the use of granulated data provides clearer influence factors and the indicator of resulting construction performance.; All the proposed methods are tested on the data collected from construction sites. The results demonstrate the feasibility and efficiency of the proposed models.