- 92 views
- 124 downloads
Analytical Approaches for Predicting Variance in Construction Productivity using Regression Methods
-
- Author / Creator
- Hasan, Md Monjurul
-
Modeling productivity entails establishing the relationship between various factors that impact construction productivity, connecting input factors with output productivity. While prefabrication facilities may limit the influence of external project factors on productivity, complexity, and variability in product design, productivity-influencing factors from the internal project environment regulate the production flow and still causes productivity to vary broadly. Such factors are numerous, and it is impossible to account for every relevant detail in the model. In addition, the developed model needs to be logically driven, transparent, easy to use, and practical to secure the trust of the practitioners. Data collection efforts for maintaining the data-driven model must be practically optimized to reduce the overall overhead cost. Therefore, a systematic, transparent, logic-driven, and quantitative approach for modeling productivity would sufficiently transform a particular set of significant input parameters into the output of corresponding productivity. Furthermore, considering those input parameters that describe the fabrication of certain products, productivity variation in connection with each input parameter of the model is attributed to customization of the fabricated product of the same type, which needs to be accounted for to estimate the likely range of the predicted productivity. In other words, besides the point value prediction of productivity, the model should also provide the variance estimate of the prediction to gauge the associated variations.
Regression methods like multiple linear regression (MLR) and the model tree (MT) have been mainstream methods for quantitatively modeling labor productivity. Both of these techniques are instrumental in generating productivity prediction models that are transparent and understandable; hence, model predictions can be trusted by decision-makers. It is important to note that even though model tree algorithms are considered as the nonlinear regression model, which combines decision trees and MLR analysis to establish complex-nonlinear relationships between variables. Such regression models are generally established based on analyzing the error terms between the predicted output and the target output without addressing the variance of the predicted output and the impact of individual input parameters on the variance. A model with high accuracy (the mean of the prediction close to the target value) but low precision (too high the variance of the prediction) would be deemed inadequate in the context of cost-estimating applications. An analytical method to account for the impact of the variability associated with each input parameter on the variation of the final output has yet to be formalized.
This research critically reviews established methods for variance analysis on commonly applied regression methods, namely MLR and model tree in cost estimating and productivity prediction for fabricated construction, as well as the impact of productivity variance on project cost budgeting. This research first proposes a novel method that integrates the error propagation theory with MLR modeling in an attempt to quantify the variance of the MLR predicted output. A metric based on the resulting variance analysis (i.e., the ratio of standard deviation over mean) for gauging the precision of the MLR model has also been proposed. Next, the variance analysis technique is used to enhance the model tree algorithm to extend its capacity to make predictions along with estimating the variance of the predicted output. A productivity modeling framework, therefore, has been formalized using the enhanced model tree algorithm to connect the unique design features of fabricated products (e.g., structural steel, prestressed concrete elements) as input with productivity as output. The productivity model’s performance has been cross-checked against the models prepared using MLR and artificial neural network (ANN) models. The enhanced model tree outperforms MLR in prediction accuracy and is preferred to ANN because (1) the variance is analytically predicted alongside the point-value output, and (2) the productivity model is explainable in terms of the reasoning logic for productivity prediction. In addition, two additional questions in connection with the variance in productivity prediction are addressed in this research, namely: (1) how the variability encoded in the fabrication productivity at work packages propagates from the work package level to the entire project level through the project network schedule and (2) how variations in project-level labor-hour are accumulated over the course of the project duration. Proposed methodology has been verified using Monte Carlo simulation and validated by conducting case studies based on fabrication projects in the real world. -
- Graduation date
- Fall 2023
-
- Type of Item
- Thesis
-
- Degree
- Doctor of Philosophy
-
- License
- This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.