Application of Data Mining Techniques for Conceptual Cost Estimation of Selected Building Projects in Addis Ababa
No Thumbnail Available
Date
2021-12
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
For project managers and decision makers, developing an accurate cost estimate in the
conceptual stage of a project is a crucial but challenging task. Different techniques and methods
have been devised and researched to accurately estimate the cost of building projects at the
preliminary stages. These methods can broadly be divided into two based on the approach they
follow. The cost –based or parametric cost modeling approach uses historical cost data and
different Data Mining techniques to develop a cost prediction model. The second method uses
a bottom-up or quantity strategy, in which data on the quantity of works is utilized to construct
quantity prediction models for each work item. These predicted quantities can then be
multiplied by their current unit rates to determine the respective costs. In this study a parametric
cost model is first developed to assess its accuracy in predicting the final cost of building
projects based on historical data collected from selected building projects in Addis Ababa. This
was then followed by doing a comparison between the cost based and quantity based
approaches by developing models for structural cost prediction as well as quantity models for
the different work items that make up the structural work (concrete, reinforcement, and
formwork). Concurrently, the study explored the effectiveness of four data mining techniques,
namely Linear Regression (LR), Decision Trees (DT), Neural Networks (ANN), and Gradient
Boosted Trees (GBT) in estimating the final and structural cost of building projects. With a
relative error of 37.05%, the ANN model was the most accurate in forecasting the final cost of
a construction project, while the GBT model performed better in predicting structural costs
with a relative error of 22.67%. For quantity estimation models, the NN model showed superior
performance for concrete and reinforcement quantity estimation with a relative errors of
16.44% and 19.32% respectively. The GBT model on the other hand performed better in
formwork quantity estimation with a relative error of 19.58%. Accordingly, the total slab area
was identified to be the most important variable for all prediction. The study indicated the
quantity based approach provides more accurate cost prediction as opposed to the cost based
approach for the case of structural cost estimation.
Description
Keywords
Data mining, Conceptual Cost Estimation, Structural Cost Estimation, Quantity Estimation, Artificial Neural Network, Gradient Boosted Trees, Linear Regression, Decision Trees