|
Activity Number:
|
191
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 1, 2016 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Korean International Statistical Society
|
|
Abstract #320047
|
|
|
Title:
|
Unsupervised Bump Hunting with Split-and-Recombine
|
|
Author(s):
|
Miriam Elman* and Jinho Park and George Tiao and Dongseok Choi
|
|
Companies:
|
Oregon Health & Science University and Inha University and The University of Chicago and Oregon Health & Science University
|
|
Keywords:
|
Bump hunting ;
split-and-recombine ;
cluster analysis ;
mode hunting
|
|
Abstract:
|
Bump hunting in machine learning seeks to distinguish anomalous data concentrations from an underlying distribution. We develop a novel approach for unsupervised bump hunting based on the split-and-recombine (SAR) procedure. The SAR was originally developed for outlier detection and adapted to clustering. This iterative two-step algorithm (1) splits data into homogenous basic sets then (2) grows these sets using a predictive measure to test whether subsets can be recombined. We compare our technique to other methods from machine learning using simulations.
|
Authors who are presenting talks have a * after their name.
