scanCP: Deep Learning–Based Changepoint Detection with Local Neural
Models
Implementation of deep learning–based changepoint detection
algorithm designed for time series with smooth local fluctuations.
The method fits localized feed‑forward neural networks to approximate the
underlying smooth component and constructs a residual‑based detector that
isolates abrupt structural changes. A fully data‑adaptive
empirical cumulative distribution function (ECDF) based thresholding
rule and refinement procedures yield accurate changepoint localization
without parametric assumptions on noise or trend structure.
| Version: |
0.1.0 |
| Imports: |
plotly, RSNNS, foreach, doSNOW, parallel, pracma, stats, magrittr, tidyr |
| Published: |
2026-05-30 |
| DOI: |
10.32614/CRAN.package.scanCP (may not be active yet) |
| Author: |
Arman Azizyan [aut, cre],
Abolfazl Safikhani [aut] |
| Maintainer: |
Arman Azizyan <arman.azizyan at gmail.com> |
| License: |
GPL-2 |
| NeedsCompilation: |
no |
| Materials: |
README |
| CRAN checks: |
scanCP results |
Documentation:
Downloads:
Linking:
Please use the canonical form
https://CRAN.R-project.org/package=scanCP
to link to this page.