I-Corps: Data-Enabled Forecasting Tools for Big Data

University Of Maryland, College Park, College Park MD

Investigators

Abstract

This research explores the ability of data-derived modeling to extract, without a priori assumptions, the inherent features of a system from its time series data. Although the essential elements of this approach to modeling were developed for geospace phenomena, it is applicable to many natural as well as social phenomena. With the increasing importance of Big Data the fundamental nature of data-derived techniques provide a new approach to the modeling of systems from their time series data. Often such systems are not readily modeled using first principles and data-derived modeling can be the only viable approach. As in the case of geospace the dynamical models can lead to data-enabled forecasting tools. Another aspect of the research is the ability to quantify the variability of the system using a new fluctuation analysis, which yields improved fluctuation exponents such as the well-known Hurst index. The modeling of dynamical behavior, prediction, forecasting and characterization of is an integral part of the emerging data-enabled science. The data-enabled forecasting tools process large data sets to extract the essential features, predict the trends and quantify the forecasting ability. A broader impact of these tools is in addressing the need of many social and commercial systems for forecasts of future trends. In financial markets, the tool could yield forecasts for stock and other instruments, quantify the reliability of the forecasts, and predict changes in the short-term trends. In natural hazards, these techniques can be used to predict extreme events such as hurricanes, floods, earthquakes, and tsunamis from the time series data. The forecasting tools are independent of pre-determined models or parameters, and thus can provide reliable analyses of extreme events in commercial (financial markets, insurance) and social (disaster planning and management) sectors. A key need of Big Data is reliable analytic tools and the proposed data-enabled tools will address this need.