Title: A Unified Framework for Cold Start Time Series Forecasting in the Presence of Missing Data
Advisors: Emily Fox and Sham Kakade
Abstract: Providing long-range forecasts is a fundamental challenge in time series modeling, which is only compounded by the challenge of having to form such forecasts when a time series has never previously been observed. The latter challenge is the time series version of the cold start problem seen in recommender systems which, to our knowledge, has not been directly addressed in previous work. Time series models are also often plagued by missing data and high-dimensionality (i.e., large collections of observed series), making them ill-suited to the typical structure of big data time series. We provide a unified framework for producing long- range forecasts even when the series has missing values or was previously unobserved; the same framework can be used to impute missing values. Key to the formulation and resulting performance is (1) leveraging repeated patterns over fixed periods of time and across series, and (2) metadata associated with the individual series. We provide an analysis of our framework on web traffic in a Wikipedia dataset.