Using Model Trees to Represent Knowhow of Experienced Estimators in Steel Fabrication Industry

To facilitate knowledge representation and transfer, this research explores the feasibility of implementing the machine learning technique of model trees in a practical application setting. In collaboration with a steel fabricator company in Canada, we tapped into the mental model of experienced estimators in the steel fabrication domain by analyzing pre-bid estimate data and evaluating the performance of model trees alongside other mainstream machine learning methods. The resulting model was validated by comparing its logic against that of the experienced estimator, demonstrating close alignment. Additionally, the model delivered reliable prediction accuracy. Our case study concludes that the technique of model trees is capable of generalizing hidden patterns and implicit relationships in the training data; to a certain degree, the model trees has generated a sufficient and explicit representation of the know-how of experienced estimators in the industry.