Preliminary Estimates of Residential Building Quantities Using Artificial Neural Networks Within Libya.

Abstract

One of the most significant challenges faced by a newly graduated engineer or anyone who has not practiced work in the cost engineering sector or even those wishing to build a building is the preliminary estimation of the building cost before setting up the schematic diagrams to know, even approximately, the suitability to which the allocated budget matches the actual cost. This research aims to use a set of different techniques to estimate these quantities in the preliminary stages based on: the quantity per unit area and statistical models through Microsoft Excel and the use of artificial neural networks through MATLAB using the feedforward neural network and using the binary sigmoid function. Eight schematic diagrams were collected with their quantities accurately estimated for the purpose of building models using the mentioned techniques and for evaluating, so that 7 schematic diagrams were used in the model building process and one plan in the testing process. As for the artificial networks method, 7 plans were used in construction and training and one for evaluating. The results showed that all techniques gave acceptable results, but the method based on the quantity per unit area was the most accurate for most items, as it reached the point of almost identicalness for many items. As for the statistical method, it was less accurate than the first method, and the last was the neural network method, which was the least accurate and was clearly greatly affected by the number of samples and the training process.