In the master's thesis, a machine learning regression model is developed to predict the quantity of processed wood from specific forest areas. The processed wood is classified into two categories: timber logs and other products such as firewood or pulpwood. This classification is based on the significant difference in the value of the wood, as timber logs have a much higher value compared to other types of wood. The model is tailored to a real-world process, where a precise inventory of trees designated for felling is first conducted, and the trees are categorized into volumetric (diameter) classes based on the diameter of the trunk. After felling, the quantity of wood is allocated into different categories, such as logs, firewood, and pulpwood. The gross volume measured before felling and the net volume measured after felling represent the central variables of the research. Additional features, such as the type of logging and the tariff class, are also included in the analysis to improve the accuracy of the predictions.
The primary goal of the thesis is the development of an accurate predictive model based on the collected harvesting data, enabling the prediction of the net volume of harvested wood ready for transport. Advanced regression models such as the multilayer perceptron regressor, linear regression, random forests, and gradient boosting regressor are used. The methodology for storing and further analyzing predictions and results using Power BI contributes to a deeper understanding and analysis of the dynamics of the wood processing industry.
|
