Enhanced Model Tree Application Framework for Developing Interpretable AI in Construction Engineering

• Author / Creator

  Naumets, Serhii

• The construction industry has been and continues to be overflown with data. Scholars have no problems dealing with this phenomenon through the incorporation of artificial intelligence (AI) methods like neural networks or random forests. However, when the time comes to practical application, the industry professionals show very little interest in these predictive models. Most of the best-performing methods are too complex and packed in "black boxes". In my view, the user’s trust in a computer program is analogous to the user’s trust in a co-worker: if there is no understanding—there is no trust, if there is no trust—there is no cooperation.

  In collaboration with a steel fabrication company in western Canada, this research investigated the cost estimation department in regards to preparing pre-bid estimates. I found that most of the estimators fall under the "baby boomer" cohort. In my view, it was imperative to capture their experience and know-how before they retire and pass it on to the next generation of engineers and managers. Another finding showed that the professionals in this company were not eager to use AI techniques. They needed something that could be easily interpreted and trusted.

  A data set sourced from this steel fabrication company was used to compare various AI algorithms and search for candidate for interpretable AI. Firstly, I identified interpretable performance metrics the meaning of which can be easily explained to a user. Secondly, these metrics were put together in a color scheme that could help to decide on the credibility of the AI model.

  As a testing case study, I used the Compressive Concrete Strength data set to illustrate that the developed framework could build an interpretable AI model in a different problem domain. Linear regressions provided by the Model Tree can serve as formula sheets to customize concrete mix or to calculate the compressive strength of concrete at a certain point of curing.

  After the comparison of Artificial Neural Network, Support Vector Machine, Random Forest, and Model Tree, the last was determined as a potential candidate to generate interpretable AI for practical applications. The enhanced M5P algorithm with three-colored performance scheme has no analogous concepts and functions in any existing software.

  As supporting material, Appendix C provides a manual of how to setup a Model Tree in WEKA and Appendix D contains the configurations of all of the discussed AI models.

• Subjects / Keywords

  • Model
  • Artificial
  • Interpretable
  • Engineering
  • M5P
  • Tree
  • Intelligence
  • Construction
  • AI

• Graduation date

  Fall 2020

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/r3-va9b-1r71

• License

  Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.