Online Program

Return to main conference page
Thursday, May 30
Machine Learning
SADM Invited Papers
Thu, May 30, 1:30 PM - 3:05 PM
Grand Ballroom I

Mean Residual Function: a Tool for Exploring Patterns in Big Data (305069)

Omid M. Ardakani, Georgia Southern University 
Majid Asadi, University of Isfahan 
Nader Ebrahimi, Northern Illinois University 
*Ehsan S. Soofi, University of Wisconsin-Milwaukee 

Keywords: Mean excess, Bayes risk, empirical distribution, functional classification, NY City Taxi data

The mean residual function (MRF) is the expected value of the excess of a random variable above a threshold. This measure is used in various applications problems in various fields. We will illustrate that MR provides a useful tool for exploring patterns in big data. An example explores patterns of the New York City's taxi trip times in various years, months, and time of the trip for the Yellow and Green taxis. We also provide some theoretical results for the Bayes risk of the MRF, defined by the expectation of the MRF with respect to a prior for the threshold. When the threshold and the variable are identically distributed the Bayes risk is the entropy functional of the survival function. We show that the standard deviation provides a tight upper bound for this Bayes risk. Attainment of the bound characterizes the exponential distribution.