Bioenergy Production and Consumption Prediction: The Best Predictors for the Best Machine Learning Models from Hundreds of Variables

The selection of variables that can be used to predict another variable is usually a challenge, considering that national and international databases contain a considerable amount of information and that the literature, in some circumstances, is unclear about the most adjusted predictors and the most accurate models. Without appropriate approaches to select the variables, there are actual risks of considering irrelevant information and ignoring important data in the prediction analysis. Artificial intelligence and, in particular, machine learning methodologies provide interesting support for identifying the most important predictors and the most accurate algorithms. In this way, this research intends to identify the most important variables to predict bioenergy production and consumption and select the most accurate models. For this, statistical information from the FAOSTAT database for the year 2023 was considered. This information was analysed considering machine learning approaches following IBM SPSS Modeler (Version 18.4) procedures. The results obtained indicate that 50% of bioenergy is produced and consumed worldwide by five countries (India, China, the United States of America, Brazil and Ethiopia) and most of this energy comes from firewood (60%). Out of a total of 456 inputs (consideration of this set of FAOSTAT variables is a novelty in the literature), bioenergy production and consumption are mainly explained by fuelwood production, with elasticities of 0.75% and 0.7%, respectively. The explanatory variable “fuelwood production” was identified from the most significant variables found by the machine learning approaches and was subsequently used as an independent variable in linear regressions. XGBoost Linear, XGBoost Tree, Linear, CHAID, Tree-AS, and C&R Tree are the most accurate models (lower relative error) for predicting bioenergy production and consumption worldwide.