TY  - CONF
TI  - Method for Classifying Wastes by Making-do Using Machine Learning
C1  - Osaka and Kyoto, Japan
C3  - Proceedings of the 33rd Annual Conference of the International Group for Lean Construction (IGLC 33)
SP  - 822
EP  - 833
PY  - 2025
DO  - 10.24928/2025/0274
AU  - Amaral, Tatiana Gondim do
AU  - Maciel, Caio César Medeiros
AU  - Junior, Marcos Paulino Roriz
AU  - Paula, Gabriella Soares de
AD  - Full Professor, Environmental and Civil Engineering Department, Federal University of Goiás (UFG), Goiânia, GO, Brazil. tatianagondim@ufg.br, orcid.org/0000-0002-9746-4025
AD  - Master Student, Program in Production Engineering, Federal University of Goiás, Brazil, caiocesar.eng@hotmail.com, orcid.org/0000-0002-6756-4068
AD  - Adjunct Professor, Faculty of Science and Technology, Federal University of Goiás (UFG), Aparecida de Goiânia, GO, Brazil, marcosroriz@ufg.br, orcid.org/0000-0003-2795-0009
AD  - Student, School of Environmental and Civil Engineering, Federal University of Goiás, Brazil, gabriella.soares@discente.ufg.br, orcid.org/0000-0002-7730-2838
ED  - Seppänen, Olli 
ED  - Koskela, Lauri 
ED  - Murata , Koichi 
AB  - The study aims to develop an automated method for classifying making-do wastes using machine learning (ML) techniques. Manual classification of these wastes is prone to inconsistencies, especially in projects with large volumes of data. The automated method makes the process more efficient and accurate. The research is classified as quantitative and empirical, with a descriptive, exploratory and experimental approach. Data was collected using the Melius quality platform and covered six high-end multi-family residential developments from two construction companies in Goiânia. The data was processed in two phases: initially, compliance was checked manually and then the data was adjusted for the ml algorithm. Preliminary results indicate that the main causes of waste are related to lack of labor (67.11%) and problems with materials (15.48%). The highest incidences of waste categories were inadequate, sequencing (26.82%) and lack of equipment (18.21%). In terms of impact, the recurrence of rework (13.37%) and lack of terminality (13.19%) stand out. The neural network model showed unsatisfactory results, with a recall of 55.4% and precision of 53.7%. The study shows the potential of machine learning, but adjustments to the models are necessary to improve their effectiveness.
KW  - Making-do waste
KW  - Machine Learning
KW  - Predictive models
KW  - Automatic classification
KW  - Neural Network.
PB  - 
T2  - Proceedings of the 33rd Annual Conference of the International Group for Lean Construction (IGLC 33)
DA  - 2025/06/02
CY  - Osaka and Kyoto, Japan
L1  - http://iglc.net/Papers/Details/2419/pdf
L2  - http://iglc.net/Papers/Details/2419
N1  - Export Date: 01 June 2025
DB  - IGLC.net
DP  - IGLC
LA  - English
ER  -