Modelo fuzzy evolutivo interpretável para predição de séries temporais no mercado financeiro.

Abstract

Financial systems encompass numerous variables, both deterministic and heuristic, ren dering them intricately complex to model mathematically. A framework was developed to address uncertainties inherent in financial markets, employing fuzzy logic and a data clustering algorithm with a focus on semantic data interpretation. Accordingly, this project aimed to construct a system capable of making precise predictions regarding the future values of financial assets, thereby enhancing prediction credibility through explainability techniques. To achieve this objective, an evolving fuzzy model was employed for non linear data representation, alongside the adaptation of the Online Elliptical Clustering (OEC) algorithm for data clustering and change point detection. The evolving fuzzy model integrates multiple linear systems within distinct ranges, termed local models, which are selected through knowledge-based methods. The adaptation of the OEC algorithm facilitates an evolving approach to semantic data representation, utilizing hyper-ellipsoids for data clustering and change point detection to update antecedents. Autoregressive models were employed to represent each local model. Implementation was conducted in Python, utilizing existing open-source libraries for data handling, cluster construction, and model representation. Ultimately, the proposed model yielded suitable predictive out comes, assessed through metrics including MAE, RMSE, MAPE, and sMAPE, effectively providing elucidations for predictions and nuances in the series behavior.