Smarter City Predictive Analytics Using Generalized Additive Models

• Activity Number: 502
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Bei Chen*+ and Mathieu Sinn
• Companies: IBM Research and IBM
• Keywords: forecasting ; applied statistics ; GAM ; model selection ; uncertainty

Abstract

Establishing efficient energy and transportation systems are key challenges for accommodating the fast-growing population living in cities. Timely and accurate forecasting of energy and transportation systems can effectively alleviate energy dependence, air pollution and traffic problems of large cities. In this talk, I will present a class of big data predictive algorithms for smarter city applications based on the Generalized Additive Models (GAMs) (Tibshirani and Hastie, 1990). The first application is short-term electricity load forecasting at various aggregation levels in the electric grid, ranging from highly aggregated series (national and regional demand), clusters of smart meters to individual buildings. The second application focuses on multi-modal transportation networks, including prediction of shared bike schemes and of urban traffic. For both applications, I will discuss in detail methods for real-time model selection from a large number of covariates and uncertainty predictions. Also, I will highlight open problems in smarter city predictive analytics for future research.