A comparative approach of methods to estimate machine productivity in wood cutting

Abstract

Forest harvesting planning requires careful analysis of the variables that influence machine productivity. This information is crucial for better decision-making. Thus, we aimed to compare models for predicting the excavator-based grapple saw productivity in wood cutting with variables from environmental data, forest inventory, and operator records. We applied Stepwise linear regression, Random Forest (RF), and Artificial Neural Networks (ANN) to estimate machine productivity (mp). Hybrid methods were also designed to perform the feature selection procedure. A Genetic algorithm (GA) was combined with RF (GA-RF), and ANN (GA-ANN). These methods were assessed according to error metrics and accuracy. Although the order of the variables’ importance changed based on these methods, the operator’s experience was the main factor in the mp behavior, regardless of the model. The work shift impacted the machine productivity, but not as significantly as the operator’s experience. The mean individual tree volume and precipitation also made a considerable contribution to the mp estimates of the GA-RF and GA-ANN models, respectively. Our findings indicate that the RF and GA-RF methods perform best and with high accuracy to estimate mp. Furthermore, we highlight that GA-RF performed a robust selection of the variables that influenced the mp behavior.