A classification method for making-do waste using Machine Learning

Abstract

The present study investigates the potential use and best-fitting model for the automated classification of wasts from making-do in construction sites, using Machine Learning techniques to reduce the labor and inconsistencies of the manual method. Given the difficulty of manually analyzing a robust textual database of non-conformities, an automated method applying Machine Learning algorithms is proposed. A total of 8,196 records were collected from the Melius Qualidade service management platform, covering twenty-one high-end multifamily projects from three construction companies in Goiânia/GO, of which 3,598 were deemed suitable for the research after filtering. The initial classification was done manually, followed by the application of nine Machine Learning algorithms using the Orange Data Mining software for testing and evaluation. Results indicated that grouping data by company yielded the best prediction accuracy, with the Neural Network model achieving a recall of up to 98.20%, making it the most effective. The study highlights that automation accelerates the classification process and improves precision and consistency in identifying wasts from making-do, significantly contributing to the optimization of quality management in construction projects.