e-AutoMFIS2: A Multi-Objective Fuzzy Model for Interpretable Multivariate Time Series Forecasting

This article introduces e-AutoMFIS2, an enhanced version of the e-AutoMFIS model, which integrates fuzzy logic and ensemble techniques for multivariate forecasting. While the initial version delivered satisfactory accuracy and interpretability, it faced challenges related to manual parameter tuning and random data subsampling, affecting its efficiency. To overcome these limitations, e-AutoMFIS2 automates key processes using the NSGA-II genetic algorithm for hyperparameter selection through multi-objective optimization. Additionally, it refines subsample selection by leveraging cross-correlation and autocorrelation, eliminating the need for exhaustive manual searches. This automation enhances efficiency while maintaining accuracy and interoperability. The model’s effectiveness was validated through case studies, demonstrating superior performance compared to its predecessor and other forecasting models. The improvements in parameter selection and subsampling reinforce e-AutoMFIS2 as a robust tool for multivariate forecasting, offering a more streamlined and effective approach to handling complex datasets.