Applications of Machine Learning in Environmetrics: Detecting Dynamic Trend-Based Clusters

• Activity Number: 249
• Type: Contributed
• Date/Time: Monday, August 1, 2016 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Xin Huang* and Iliyan R. Iliev and Lyubchich Vyacheslav and Alexander Brenning and Yulia R. Gel
• Companies: The University of Texas at Dallas and The University of Texas at Dallas and University of Maryland Center for Environmental Science and University of Jena and The University of Texas at Dallas
• Keywords: Dynamic clustering ; Space-time data mining ; Trend synchronism ; Trend detection

Abstract

Many existing methods for detection of joint trends and other patterns in space-time data are based on grouping observations simply by geographical proximity. Such clustering is often static and thus does not account for changes in space-time data distribution. Moreover, a number of clusters is typically fixed a-priori. To relax these restrictions, we present a flexible dynamic clustering approach for space-time processes based on adaptation of streaming-data algorithms. We propose a new data-driven method for selecting optimal clustering parameters using V-fold cross validation. We evaluate our approach using synthetic data and socio-environmental case studies.