Multivariate time series forecasting has evolved significantly with the adoption of techniques based on artificial intelligence. Traditional methods such as regression trees provided robust and efficient solutions, while modern deep learning models like Long Short-Term Memory networks, Convolutional Neural Networks and Graph Neural Network architectures excel at capturing complex temporal dependencies. However, these advancements incur high computational resource demands and increased energy consumption. This challenge has spurred interest in developing sustainable forecasting approaches under the principles of Green Artificial Intelligence. The study presents a comparison of tree-based model techniques with deep learning models, using Bayesian tests to evaluate both prediction error and computational time in multivariate time series forecasting. The analysis clearly outlines the trade-offs between accuracy and efficiency, emphasizing the importance of implementing environmentally responsible forecasting methods.