Learning Penalty for Changepoint Detection via Automatic Feature Extraction
Organizations: School of Informatics, Computing, and Cyber Systems, Northern Arizona University, S San Francisco, Flagstaff, 86011, Arizona, USA. · Département d’informatique, Université de Sherbrooke, Sherbrooke QC J1K 2R1, Quebec, Canada.
Abstract
Changepoint detection identifies significant shifts in data sequences, making it important in areas like finance, genetics, and healthcare. The Optimal Partitioning algorithms efficiently detect these changes, using a penalty parameter to limit the changepoints count. Determining the optimal value for this penalty can be challenging. Traditionally, this process involved manually extracting statistical features, such as sequence length or variance to make the prediction. This study proposes a novel approach that uses recurrent networks to learn this penalty directly from raw sequences by automatically extracting features. Experiments conducted on 20 benchmark genomic datasets show that this novel method generally outperforms traditional ones in changepoint detection accuracy.
Figures & tables
| Recurrent core: | |||||
| sequence feature | used in | ||||
| length | 1 | ( Schwarz, 1978 ; Akaike, 1974 ; Lavielle, 2005 ; Drouin et al., 2017 ; Barnwal et al., 2022 ; Hocking et al., 2013 ; Hocking and Srivastava, 2023 ; Nguyen and Hocking, 2025b ) | |||
| mean | 1 | 1 | 0 | ( Drouin et al., 2017 ; Barnwal et al., 2022 ; Hocking et al., 2013 ) | |
| min | 1 | 1 | ( Drouin et al., 2017 ; Barnwal et al., 2022 ; Hocking et al., 2013 ) | ||
| max | 1 | 1 | ( Drouin et al., 2017 ; Barnwal et al., 2022 ; Hocking et al., 2013 ) | ||
| sum | 1 | 1 | 0 | ( Drouin et al., 2017 ; Barnwal et al., 2022 ; Hocking et al., 2013 ) | |
| Dataset | Seq. length | Size | Source |
|---|---|---|---|
| systematic | 0000 66 - 5937 | 3418 | |
| detailed | 0000 25 - 5937 | 3730 | GitHub |
| cancer | 0000 39 - 43628 | 826 | |
| ATAC_JV_adipose | 105716 - 11.50M | 465 | |
| CTCF_TDH_ENCODE | 166431 - 11.09M | 182 | |
| H3K27ac-H3K4me3_TDHAM_BP | 000 275 - 0 4.33M | 2008 |
| Model | Regularization | Language | Package | Paper |
| linear | L1 | R | ( Hocking, 2024 ) | Hocking et al. (2013) |
| MMIT | tree architecture | Python/Cpp binding | ( Drouin, 2017 ) | Drouin et al. (2017) |
| AFT_XGboost | tree architecture | Python | ( Chen et al., 2025 ) | Barnwal et al. (2022) |
| constant | none | Python/Cpp binding | ( Drouin, 2017 ) | Baseline |
| MLP | hidden layer number hidden layer sizes | Python | Additional | |
| RNN LSTM GRU | hidden layer number hidden state sizes | Python | Proposed |
Change-Point Detection
Change-point detection identifies when the statistical behavior or underlying regime of a sequence changes.
See all 89 papers