Time series forecasting from partial observations via Non-negative Matrix Factorization

Abstract

In modern time series problems, one aims at forecasting multiple time series with possible missing and noisy values. In this paper, we introduce the Sliding Mask Method (SMM) for forecasting multiple nonnegative time series by means of nonnegative matrix completion: observed noisy values and forecast/missing values are collected into matrix form, and learning is achieved by representing its rows as a convex combination of a small number of nonnegative vectors, referred to as the archetypes. We introduce two estimates, the mask Archetypal Matrix factorization (mAMF) and the mask normalized Nonnegative Matrix Factorization (mNMF) which can be combined with the SMM method. We prove that these estimates recover the true archetypes with an error proportional to the noise. We use a proximal alternating linearized method (PALM) to compute the archetypes and the convex combination weights. We compared our estimators with state-of-the-art methods (Transformers, LSTM, SARIMAX...) in multiple time series forecasting on real data and obtain that our method outperforms them in most of the experiments.

References in corpus (9)

• Scikit-learn: Machine Learning in Python
• The Why and How of Nonnegative Matrix Factorization
• Inertial Proximal Alternating Linearized Minimization (iPALM) for Nonconvex and Nonsmooth Problems
• Factoring nonnegative matrices with linear programs
• On the Geometric Interpretation of the Nonnegative Rank
• Introduction to Nonnegative Matrix Factorization
• Structured low-rank matrix completion for forecasting in time series analysis
• A Review of Nonnegative Matrix Factorization Methods for Clustering
• Intersecting Faces: Non-negative Matrix Factorization With New Guarantees