Tensor extrapolation: Forecasting large-scale relational data

Tensor extrapolation attempts to forecast relational time series data using multi-linear algebra. It proceeds as follows. Multi-way are arranged in the form of tensors, ie multi-dimensional arrays. decompositions then used retrieve periodic patterns data. Afterwards, these serve input for methods. However, previous approaches tensor limited preselected and binary To permit automatic forecasting, paper at hand connects state-of-the-art with a general class state-space models. Moreover, it highlights need preprocessing settings real-valued In doing so, enables data-driven model selection estimation large-scale forecasting problems. Numerical experiments show effectiveness proposed method identifying relevant underlying demonstrate its superiority over established methods terms accuracy.