A Novel Approach to Predict Moisture Content on Wood Using AI

Wood is one of the most used raw materials for the manufacture of various products. When it comes to industrial processes, it is necessary to optimize its use as much as possible. A key point for this optimization is the control of the of moisture content, because wood logs gradually start to lose water when stored in open air. Such water loss is directly reflected in weight and defects, such as warping and cracks, which impair the production processes causing irreparable losses of raw material. Given this context, this work develops a method based on Machine Learning to predict the variation in the weight of wood logs caused by moisture loss, without the need to purchase equipment or increase the cost of labor, using data that companies in the sector already have at their disposal. Several models were developed, with different algorithms, hyperparameter configurations and data distributions. These models were subjected to different evaluations. As an empirical evaluation, the metric chosen was the weighted balanced F-measure. The results showed that the model developed using the Random Forest classifier, classes balanced by the SMOTE algorithm, and hyperparameters adjusted with best tuning obtained the best classification performance. In the statistical analysis, the Align-Friedman and post hoc Holm tests were used. Considering the model based on Random Forest as a control method, the results showed that the algorithms Random Forest and Gradient Boosting can be considered statistically equivalent in most of the scenarios tested. Thus, a model based on Random Forest is suggested for industrial use, thus enabling better use of wood logs in production processes and helping decision-making in the industrial chain in which wood is inserted.