Detecting fraudulent whiplash claims by support vector machines

Steinn Gudmundsson*, Gudny Lilja Oddsdottir, Thomas Philip Runarsson, Sven Sigurdsson, Eythor Kristjansson

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

7 Citations (Scopus)

Abstract

A new method is proposed for detecting fraudulent whiplash claims based on measurements of movement control of the neck. The method is noninvasive and inexpensive. The subjects track a slowly moving object on a computer screen with their head. The deviation between the measured and actual trajectory is quantified and used as input to an ensemble of support vector machine classifiers. The ensemble was trained on a group of 34 subjects with chronic whiplash disorder together with a group of 31 healthy subjects instructed to feign whiplash injury. The sensitivity of the proposed method was 86%, the specificity 84% and the area under curve (AUC) was 0.86. This suggests that the method can be of practical use for evaluating the validity of whiplash claims.

Original languageEnglish
Pages (from-to)311-317
Number of pages7
JournalBiomedical Signal Processing and Control
Volume5
Issue number4
DOIs
Publication statusPublished - Oct 2010

Other keywords

  • Classification
  • Insurance fraud
  • Support vector machines
  • Time series
  • Whiplash

Fingerprint

Dive into the research topics of 'Detecting fraudulent whiplash claims by support vector machines'. Together they form a unique fingerprint.

Cite this