NFISiS: new perspectives on fuzzy inference systems applied to renewable energy, finance, and cryptocurrency

Abstract

Fuzzy inference systems are a special class of machine learning models known for balancing accuracy with interpretability. The two main types of fuzzy inference systems are Mamdani and Takagi-Sugeno-Kang. While Mamdani models prioritize interpretability, Takagi-Sugeno-Kang models achieve higher accuracy by approximating nonlinear systems through a collection of linear subsystems. However, designing fuzzy rules lacks a standardized method, often leading to models with excessive hyperparameters, high complexity, and no direct control over the number of rules. To address such challenges, this work introduces two new data-driven approaches for designing Mamdani and Takagi-SugenoKang rules. Among the advantages of the proposed fuzzy inference systems with the new mechanism for rule generation, the following can be highlighted: reduced complexity, fewer hyperparameters, enhanced interpretability, and direct control over the number of final rules. Additionally, feature selection techniques, including genetic algorithms and ensemble methods, are integrated to improve the models’ ability to handle large datasets, optimize performance, increase interpretability, and prevent overfitting. The introduced models result in the NFISiS, a new collection of data-driven fuzzy inference systems for time series forecasting. The proposed models are evaluated using benchmark time series, renewable energy, financial, and cryptocurrency datasets in terms of errors and the number of rules. Their performance is compared against state-of-the-art machine learning models, including classical approaches, deep learning architectures, and rule-based evolving Fuzzy Systems. The results indicate that the proposed models effectively handle complex, non-stationary datasets, such as those in finance and cryptocurrency. All proposed models are available as a Python package, which can be installed via pip: pip install nfisis(https: //pypi.org/project/nfisis). Furthermore, the source code used for the simulations are available at https://github.com/kaikerochaalves/nfisis_thesis.git.