Machine learning applied to the prediction of harvester maintenance employed in Eucalyptus harvesting

• Grant number: 22/06374-9
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: August 01, 2022
• End date: July 31, 2023
• Field of knowledge: Agronomical Sciences - Forestry Resources and Forestry Engineering - Techniques and Operations in Forestry
• Principal Investigator: Danilo Simões
• Grantee: Valier Augusto Sasso Júnior
• Host Institution: Faculdade de Ciências Agronômicas (FCA). Universidade Estadual Paulista (UNESP). Campus de Botucatu. Botucatu , SP, Brazil

Abstract

The accurate maintenance planning of self-propelled forest machines employed in wood harvesting can be carried out through the approach of statistical methods of machine learning, as it provides detailed analyzes from the processing of large volumes of data. That said, the objective will be to verify if the models generated by machine learning by classification, will reach precision for the prediction of the maintenance of the self-propelled forest machine harvester in planted Eucalyptus forests. The study will be developed from data from wood harvesting carried out in the Midwest region of the state of São Paulo, Brazil. The harvesting system will be cut-to-length and will involve the activities of cutting trees, debarking and stacking the wood, using the self-propelled forest machine harvester, brand Komatsu - model PC200F. The database will be structured from a dataset, with different attributes and, consequently, different instances. Subsequently, there will be the construction of predictive models related to maintenance. Afterwards, the data will be divided into training and test sets. Then there will be the balancing of the training set and the supervised learning of classification will be applied. 18 learning algorithms will be employed, based on decision tree, linear regression, K-Nearest Neighbors and Support Vector Machine, in default mode, to the set. Those that achieve the best prediction performances will proceed to tune, ensemble, blend and stack processes to adjust the hyper parameters and provide an increase in the predictive performance of the models, applying them to the test set. The prediction performance of the models will be evaluated in both sets, with metrics Area under the ROC Curve, Accuracy, Matthews Correlation Coefficient, F1-Score and Precision.(AU)