August 2005

RESCUE research aims at forecasting movements of people through time and space

As a weather forecast is valuable to someone planning an outdoor activity, imagine the value of people-forecasting tools that accurately predict the location and movement of people to those in emergency situation assessment, response, planning and simulation. Over the past year, RESCUE researchers led by Professors Rina Dechter and Padhraic Smyth have been developing probabilistic models for learning and predicting where people are and where they are traveling.

In this research, a model was developed to aid in obtaining reliable estimates of the number of people traveling from one region to other (for example, UCI to Irvine Valley College) at any given time for an entire urban area. Currently, the field of transportation science estimates this number by relying upon out-dated pencil-and-paper travel surveys and sparsely distributed sensors like loop-detectors along major intersections. We view the increasing proliferation of powerful mobile computing devices like GPS-enabled cars and cell phones as an opportunity to remedy this situation. If even a small sample of the traveling public agreed to collect their travel data and make that data publicly available, transportation management systems could significantly improve their operational efficiency. An important starting point is to develop an efficient formulation for learning and inferring individual traveler routines like a traveler's destination and his/her route to destination. Our probabilistic model solves this problem by predicting an individual's common origins/destinations and his/her usual routes with a very high accuracy. You can read the technical contributions in the paper "Modeling Transportation Routines using Hybrid Dynamic Mixed Networks," available at http://www.ics.uci.edu/~csp/r124.pdf .

Efforts are underway to develop a modeling framework that can infer how many people are (or were) on the UCI campus (or in a specific area or building on campus) at any given time (in the past, currently, or in the future).  The framework will fit a statistical model to historical data, taking into account known systematic effects such as calendar and time of day.