دانلود رایگان مقاله آنالیز بیزی سری زمانی با استفاده از روش محاسبات دانه

ABSTRACT

The soft computing methods, especially data mining, usually enable to describe large datasets in a humanconsistent way with the use of some generic and conceptually meaningful information entities like information granules. However, such information granules may be applied not only for the descriptive purposes, but also for prediction. We review the main developments and challenges of the application of the soft computing methods in the time series analysis and forecasting, and we provide a conceptual framework for the Bayesian time series forecasting using the granular computing approach. Within the proposed approach, the information granules are successfully incorporated into the Bayesian posterior simulation process. The approach is evaluated with a set of experiments on the artificial and benchmark real-life time series datasets.

6. Conclusion

In this paper we have reviewed the main challenges and the recent developments of incorporating the soft computing techniques into the statistical time series analysis and forecasting. We have introduced the Bayesian Granular Computing (B-GC) method for Time Series Forecasting, that aims at the identification of the a priori model distributions with the use ofthe information granules mined from the time series datasets with soft computing techniques. One of the main advantages of the proposed solution is its human-centricity, being a result of including the imprecise information granules. This is especially importantin practice for experts involved in the forecasting process for large datasets. The data mining and classification methods employed in the work provide useful information granules. It is shown that the considered linguistic summaries constitute intuitive and very informative knowledge,and may successfully support the Bayesian time series forecasting process. The performance ofthe approach is illustrated with the artificial and real-life datasets. The simulations confirm that the construction of the prior model probabilities with support of the selected granular computing techniques may lead to the increase of forecast accuracy compared to the uniform Bayesian averaging and the competitive benchmark methods. The numerical results prove that the proposed approach delivers very competitive results, not only in terms of its human-centricity, but also in terms ofthe forecasting accuracy. Future research assumes the incorporation of other types of information granules and probabilistic models into the forecasting framework. We conclude that the intelligent combination of the granular computing achievements and the Bayesian modeling is a promising direction for the future research on the time series analysis and forecasting.