Modeling and Forecasting Time Series of Compositional Data: A Generalized Dirichlet Power Steady Model

摘要: This paper presents GDPSM a power steady model (PSM) based on generalized Dirichlet observations for modeling and predicting compositional time series. The model’s unobserved states evolve according to the conjugate prior distributions. observations’ distribution is transformed into set of Beta distributions each which re-parametrized as unidimensional in its exponential form. We demonstrate that dividing problem multiple smaller problems leads more accurate predictions. evaluate this with web service selection application. Specifically, we analyze proportions quality classes are assigned services interactions. Our compared another PSM assumes observations. experiments show promising results terms precision errors standardized residuals.