This paper proposes a novel general Gaussian process framework for change point detection in time series. A change is defined as an alteration of the hyperparameters of the Gaussian process prior. We formulate the problem in the context of statistical hypothesis testing and demonstrate its performance over both offline and online multiple change points detection. The approach provides theoretical guaranties of its performance. An important advantage of the developed framework is that it does not need a detailed specification of the time series function before and after a change. The hyperparameter values are learned from the data. Comprehensive evaluation of the proposed framework is provided both on synthetic and real data and its efficiency is demonstrated.
Figures & tables
Figure 1: Empirical normalised histogram of the test likelihood-based statistic with the corresponding analytical χ2 distribution.
Figure 2: Minimum power of the GP-OLCDT within the area A of hyperparameters for the alternative H1 as a function of the variance σ02 of the additive noise.
Figure 3: Power of the GP-OLCDT as a function of the hyperparameters for the alternative H1 . The null hypothesis hyperparameters are plotted as a diamond. Triangular points correspond to the values of the alternative hyperparameters from the four considered scenarios.
Figure 4: Sample observations generated under four scenarios by the proposed generative model.
Scenario
Length scale, ℓ1
Signal variance, σK12
Power of the GP-OLCDT with known hyperparameters
1
1
1
0.99
2
20
1
0.06
3
3
0.3
0.10
4
3
4
0.95
Table I: Scenarios characteristics for data simulated by the proposed generative model. The null hypothesis hyperparameters are: ℓ0=3 and σK02=1
Figure 5: Cdfs of the test statistic for the GP-OLCDT with the known hyperparameter vector ϑ under the H0 and H1 hypotheses. The vertical lines correspond to αstat/2 and 1−αstat/2 quantiles of the the test statistic distribution under the null hypothesis H0 , the significance level αstat is set to 0.01 . These quantiles are used as lower and upper bounds for the test statistic to make a decision of rejecting H0 . The power of the test values are also marked in the plots.
Method
Mean time between false alarms
Mean delay for detection
Missed detection rate
Power
Scenario 1
GP-OLCDT with known ϑ0
157.67
1.31
0.00
0.99
GP-OLCDT with batch estimated ϑ0
164.53
1.24
0.00
0.99±0.00
GP-OLCDT with sliding estimated ϑ0
48.19
1.56
0.01
0.99±0.01
GP-OLCDT with adaptively estimated ϑ0
87.54
6.48
0.08
0.99±0.01
GP-LRT
195.95
0.60
0.04
—
Table II: Change point detection performance on data simulated by the proposed generative model
Figure 6: Sample observations generated under 2 scenarios by the GP-BOCPD generative model.
Method
Mean time between false alarms
Mean delay for detection
Missed detection rate
“Easy” scenario
GP-OLCDT with batch estimated ϑ0
151.37
6.59
0.05
GP-OLCDT with sliding estimated ϑ0
94.48
4.25
0.24
GP-OLCDT with adaptively estimated ϑ0
71.86
9.42
0.16
GP-GLRT
181.19
0.87
0.30
GP-BOCPD
197.85
10.13
0.25
Table III: Change point detection performance on data simulated by the GP-BOCPD model
Figure 7: Honey bee dance data. is a sample frame of the original video. is a scheme of dance phases.
Figure 8: Change point detection on the honey bee dance data. Two out of total six sequences are presented. x and y coordinates are artificially separated for plotting to increase readability. Vertical lines correspond to true change points. Crosses represent detections by different algorithms.
Method
Mean time between false alarms
Mean delay for detection
Missed detection rate
Time (sec)
Sequence 1
GP-OLCDT with sliding estimated ϑ0
17.05
2
0.37
59.94
GP-OLCDT with adaptively estimated ϑ0
23.91
2.79
0.26
2.63
GP BOCPD
25.5
12
0.84
18.08
CUSUM
12.31
2.11
0.05
3 e −3
Contrast
18.21
3.4
0.21
7 e −3
Table IV: Change point detection performance on the bee dance data
Statistical change point (CP) detection methods typically rely on likelihood-based inference and ignore contextual information about plausible CP locations beyond the observed sequence. Although informative priors provide a natural way to incorporate such information, general and computationally efficient methods for doing so are lacking, especially for multiple CP detection. To address this gap, we propose a prior-informed CP detection algorithm (Pi-Change) that incorporates prior information on CP locations through a time-varying penalty term. We prove that the proposed penalty can be embedded in the Pruned Exact Linear Time framework while preserving the dynamic programming recursion and pruning rule required for efficient multiple CP detection. Across simulation studies and three time-series applications, Pi-Change discourages spurious CPs unsupported by prior information, remains robust to prior misspecification, and improves detection accuracy. More broadly, Pi-Change extends multiple CP detection beyond purely data-driven fitting by incorporating partial prior knowledge in a computationally efficient and interpretable way. It is particularly useful when CPs arise from heterogeneous mechanisms or are associated with known external events, helping quantify the delay between an event and the resulting structural change.
Jonathon Jacobs, Shanshan Chen
Department of Biostatistics Virginia Commonwealth University Richmond, VA, USA
This paper introduces a distribution-free framework for constructing post-detection confidence sets for changepoints after stopping a sequential change detection procedure. It is well known that conformal test martingales can be used to sequentially detect changes in distribution, but by themselves provide no inference for the time at which a proclaimed change occurred. Past work on post-detection inference requires pre- and post-change classes of distributions to be known, but this paper accomplishes localization of the changepoint without any distributional assumptions. We establish finite-sample coverage guarantees (conditional on correct detection). We provide non-asymptotic bounds on the conditional expected size of the confidence sets. Under suitable asymptotic regimes, we prove that the conditional expected size of the confidence set remains uniformly bounded and demonstrate strong empirical performance on simulated and real data. To the best of our knowledge, this is the first general distribution-free framework for sequential changepoint localization with valid post-detection coverage.
Aytijhya Saha, Aaditya Ramdas
Massachusetts Institute of Technology · Carnegie Mellon University
Skchange is an open-source Python library for detecting structural changes in time series. It implements modern change detection algorithms within a unified and extensible framework. The algorithms are modular and composable, and they include changepoint search methods based on both cost minimisation and statistical tests. Key features include the detection of anomalous segments in addition to changepoints; theoretically well-founded fast and approximate search methods; theoretically well-founded algorithms for high-dimensional data, covering settings where either few or many features change simultaneously; utilities for automatic and data-driven penalty calibration, which balances false alarms against missed detections; and a large collection of built-in costs and statistical tests. The design follows established scikit-learn conventions to streamline both user and contributor experience, and Numba is used extensively to achieve high computational performance. Source code and documentation are available at https://github.com/NorskRegnesentral/skchange.
Martin Tveten, Johannes Voll Kolstø, Per August Jarval Moen
Department of Statistics and Machine Learning, Norwegian Computing Center · Department of Mathematics, University of Oslo