École Nationale Supérieure de
Techniques Avancées

Résumé

This report contains a study of a change point detection by direct approximation of -relatice density ratio. We present a non-parametric, multi-dimensional change point detection method which has a generally satisfactory performance on any in put time series, especially for ARMA time series,then realize the algorithm on Python and compare it with existing methods. In this article, we use articial and real life data to test this detection method. A basic introduction of change point is presented at the beginning of this report. Then, a relative Pearson divergence via least-squares relative density ratio approximation is introduced. After that, the numerical experiments and caparison with other methods are shown as the third part. Finally, we make the conclusion and present problems and further work to do.