Dr. Mehdi Maadooliat will share “Forecasting Functional Time Series with Rfssa: An R Package for Functional Singular Spectrum Analysis,” for a colloquium presented jointly by the Departments of Computer Science and Mathematical and Statistical Sciences. The seminar will be held on Wednesday, Feb. 4, at 1 p.m. in Cudahy Hall 401.
In this work, Maadooliat introduces two novel algorithms for nonparametric forecasting of functional time series (FTS) data. These methods extend functional singular spectrum analysis (FSSA), enabling the decomposition of FTS into key components such as trends, periodicities and noise. The multivariate FSSA (MFSSA) algorithm is designed to handle multivariate functional time series (MFTS), accommodating variables observed over different dimensional domains. This allows for the joint decomposition of functional curves and images within a unified framework. The Rfssa R package provides a fast and user-friendly implementation of these FSSA-based techniques. It is flexible enough to handle covariates observed across different dimensional domains, facilitating joint analyses of smoothed curves and image data. The package is optimized for efficiency, leveraging RcppEigen, RSpectra and custom C++ code to ensure high performance. In addition, Rfssa includes a Shiny web application that offers an intuitive graphical user interface for applying FSSA to real or simulated FTS/MFTS data. Overall, Rfssa enables practitioners to apply advanced FSSA techniques to support data-driven decision making across a wide range of applied domains.
