Advancing Stock Price Index Forecasting Based on Hybrid Picture Fuzzy Time Series Model
| Autoři | |
|---|---|
| Rok publikování | 2025 |
| Druh | Článek v odborném periodiku |
| Časopis / Zdroj | INTERNATIONAL JOURNAL OF FUZZY SYSTEMS |
| Fakulta / Pracoviště MU | |
| Citace | |
| www | https://link.springer.com/article/10.1007/s40815-025-02146-2 |
| Doi | https://doi.org/10.1007/s40815-025-02146-2 |
| Klíčová slova | Picture fuzzy set; Fuzzy c-means clustering; Aggregation operator; Rules-based defuzzification; Stock market forecasting |
| Přiložené soubory | |
| Popis | Financial markets have a profound impact on societal well-being, influencing household net financial wealth, particularly pension-related assets. Events like as the 2007 financial crisis, Covid-19 outbreak, the Russia-Ukraine war, and oil price shocks, also contribute to extreme price fluctuations. These wealth losses not only affect households' economic behavior (e.g., employment, retirement planning, consumption, and investment), but also have implications for social behavior (e.g., intergenerational transfers) and mental health. Time series analysis is a powerful tool for understanding the dynamics of stock prices over time and predicting future behavior. However, the inherent uncertainty and ambiguity in time series data pose challenges for traditional statistical models. In order to address this specific problem, this study proposes an Artificial Intelligence (AI)-driven approach utilizing fuzzy time series modeling integrated with fuzzy clustering, information granules, picture fuzzy sets and new defuzzification rules. The linguistic values of market historical data are described by capturing the positive, neutral, and negative aspects of each observation through the application of picture fuzzy sets. This AI-enhanced approach incorporates a new method for partitioning the reorganized universe of discourse into intervals using the fuzzy c-means clustering and data granulation, and a novel membership function, composed of three Gaussian functions is defined to assign picture fuzzy memberships to each interval. Additionally, the proposed model employs a picture fuzzy weighted aggregation operator to aggregate the membership information across a multiple picture fuzzy sets, and use a rule-based method for defuzzifying the picture fuzzy sets to obtain crisp forecasts. The proposed model is evaluated on the TAIEX dataset and compared with several existing fuzzy time series prediction methods in terms of standard accuracy measures. The results demonstrated that the proposed approach outperforms existing techniques, providing more accurate and reliable forecasts. Furthermore, the model highlights the potential to offer more information and insights for decision-making and analysis in financial contexts through multivariate picture fuzzy modeling. |
| Související projekty: |