Nonparametric Splitting Algorithm for Detecting Structural Changes in Predictive Relationships

Research output: Chapter in Book/Report/Conference proceedingConference contributionScientificpeer-review

Researchers

Research units

Abstract

The problem of detecting structural changes in a regression study has become crucially important in a wide variety of fields, since data generating processes in a real world are usually unstable. Taking into account the fact that relationships within observed data are often in a continuous flux, it can be challenging to make any distributional assumptions. In the current paper, we propose a new nonparametric technique which allows estimation of an unknown number of structural change points in multivariate data having univariate response. The Nonparametric Splitting algorithm is a heuristic smart search for relationship changes based on a consequential division of the data into smaller parts. The approach utilizes a nonparametric change point test to find narrow regions of change locations. Our preliminary experiments are promising and suggest potential for the high efficiency and prediction accuracy of the introduced method.

Details

Original languageEnglish
Title of host publicationProceedings of the International Conference on Compute and Data Analysis (ICCDA)
Publication statusPublished - May 2017
MoE publication typeA4 Article in a conference publication
EventInternational Conference on Compute and Data Analysis - Lakeland, United States
Duration: 19 May 201723 May 2017

Conference

ConferenceInternational Conference on Compute and Data Analysis
Abbreviated titleICCDA
CountryUnited States
CityLakeland
Period19/05/201723/05/2017

    Research areas

  • change point, structural changes, predictive relationships

ID: 15931728