A process for the development and evaluation of preliminary construction material quantity estimation models using backward elimination regression and neural networks

During the early stages of a project, it is beneficial to have an accurate preliminary estimate of its cost. One way to make those estimates is by determining the amount of construction material quantities that are required and then multiplying the estimated construction material quantities by the corresponding unit cost. One advantage of making estimates in this way is that it allows for the segregation of quantities and costs. This way they can be updated separately as new information becomes available. They can also be tracked separately allowing decision makers to make better decisions about the project during its conceptual phase. There are several techniques that can be used to develop estimation models. The most common include regression analysis and artificial intelligence, such as neural networks. Work has been done, however, in a non-standardized way, leaving practitioners without guidance as to how to develop and evaluate models for their specific purposes. This can be seen in particular in the many different types of metrics used for the evaluation of models. The goal of the work presented in this article was to create a process to (1) develop models to be used to prepare preliminary estimates of construction material quantities taking into consideration the available data during the early stages of a project, and (2) evaluate the developed models using the Akaike information criterion. The proposed process is illustrated with an example in which data from 58 storage buildings was used to develop models to estimate the amount of concrete and reinforcement required using backward elimination regression analysis and neural network techniques. The developed models were then evaluated using a second-order correction Akaike information criterion (AICc) to select the most accurate model for each construction material quantity. The proposed process is useful for practitioners in need of developing robust estimation models in a consistent and systematic way, and the AICc metric proved to be effective for selecting the most accurate models from a set.