High-dimensional changepoint detection via a geometrically inspired mapping

By detecting the changing trend, Change point detection (CPD) can describe the underlying behavior of the system with time series (e.g., equipment failure detection, auxiliary medical diagnosis, and climate change detection, etc.). However, in the current big data environment, how quickly obtaining time-series data dependencies and accurately detecting change points is still a challenge. We propose Sailfish, an unsupervised change point detection framework based on the Gaussian process for time series data. In comparison with existing CPD designs, Sailfish has two novel features: 1) using the deep Gaussian process as a hidden variable transformer and 2) integrating the Gaussian process into a bi-LSTM cell for capturing past and future embedded feature trends. Our extensive experiments show that the Sailfish significantly outperforms five state-of-the-art CPD methods with a faster speed and higher accuracy on three public real-time-series datasets. Especially on the large volume of datasets, Sailfish can achieve up to a 60% reduction in training time and 13% F1-score improvement with little overhead compared to the state-of-the-arts.KeywordsChange point detectionGaussian processBi-Long short term memoryBayesian neural network

Change points are abrupt variations in time series data. Such abrupt changes may represent transitions that occur between states. Detection of change points is useful in modelling and prediction of time series and is found in application areas such as medical condition monitoring, climate change detection, speech and image analysis, and human activity analysis. This survey article enumerates, categorizes, and compares many of the methods that have been proposed to detect change points in time series. The methods examined include both supervised and unsupervised algorithms that have been introduced and evaluated. We introduce several criteria to compare the algorithms. Finally, we present some grand challenges for the community to consider.

Change point detection (CPD) for multi-agent systems helps one to evaluate the state and better control the system. Multivariate CPD methods solve the d × T time series well; however, the multi-agent systems often produce the N × d × T dimensional data, where d is the dimension of multivariate observations, T is the total observation time, and N is the number of agents. In this paper, we propose two valid approaches based on higher-order features, namely, the Betti number feature extraction and the Persistence feature extraction, to compress the d-dimensional features into one dimension so that general CPD methods can be applied to higher-dimensional data. First, a topological structure based on the Vietoris-Rips complex is constructed on each time-slice snapshot. Then, the Betti number and persistence of the topological structures are obtained to separately constitute two feature matrices for change point estimates. Higher-order features primarily describe the data distribution on each snapshot and are, therefore, independent of the node correspondence cross snapshots, which gives our methods unique advantages in processing missing data. Experiments in multi-agent systems demonstrate the significant performance of our methods. We believe that our methods not only provide a new tool for dimensionality reduction and missing data in multi-agent systems but also have the potential to be applied to a wider range of fields, such as complex networks.

Generally, Change Point (CP) detection is time-consuming, especially from large-scale time series. In this paper, a fast method of CP detection is proposed based on Haar Wavelet (HW) and Binary Search Tree (BST), named HWBST. In this method, by multi-level HW, a Binary Search Tree, termed BSTcD, is constructed from a diagnosed time series, and two binary search criteria are introduced to detect abrupt change from root to leaf nodes in BSTcD. Then, the sensitivity and accuracy of HWBST are analyzed and evaluated on the simulated and Electrocardiogram (ECG) time series. The results show that HWBST has better performance than HW, KS, and T statistic methods, in terms of computation time, error, accuracy etc.

Many modern applications require detecting change points in complex sequential data. Most existing methods for change point detection used in streaming data settings are unsupervised and, as a consequence, lack any information regarding what kind of changes we want to detect or if some kinds of changes are safe to ignore. This often results in poor change detection performance, as many spurious changes are falsely detected as change points. We present a novel change point detection framework that uses a small number of labeled change point instances as supervision for learning a ground metric such that Sinkhorn divergences can be then used in two-sample tests on sliding windows to detect change points. Our method can be used to learn a sparse metric which can be useful for both feature selection and interpretation in high-dimensional change point detection settings. Experiments on simulated as well as real world sequences show that our proposed method can substantially improve change point detection performance over existing unsupervised change point detection methods using only few labeled change point instances.

Automatic sensing devices and computer systems have been widely adopted by the automotive manufacturing industry, which are capable to record machine status and process parameters nonstop. While a manufacturing process always has natural variations, it is crucial to detect significant changes to the process for quality control, as such changes may be the early signs of machine faults. This motivates our study on change-point detection methods for automotive manufacturing. We aim at developing a systematic approach for detecting process changes retrospectively in complex, nonstationary data. The proposed approach consists of nonparametric change-point detection, alarm generation based on change-point estimations, and performance evaluation against historical maintenance records. For change-point detection, three nonparametric methods are suggested—least absolute shrinkage and selection operator (LASSO), thresholded LASSO, and wild binary segmentation (WBS). Multiple decision rules are proposed to determine how to generate alarms from change-point estimations. Numerical studies are conducted to demonstrate the performance of the proposed systematic approach. The different change-point detection methods and different decision rules are evaluated and compared, with scenarios for choosing one set of change-point detection method and decision rule over another combination identified. It is shown that LASSO and thresholded-LASSO outperform WBS when the shift size is small, but WBS produces a smaller false alarm rate and handles the clustering of changes better than LASSO or thresholded LASSO. Data from an automotive manufacturing plant are used in the case study to demonstrate the proposed approach. Guidelines for implementation are also provided.

We develop an approach for water quality time series monitoring and contamination event detection. The approach combines affine projection algorithms and an autoregressive (AR) model to predict water quality time series. Then, we apply online change-point detection methods to the estimated residuals to determine the presence, or not, of contamination events. Particularly, we compare the performance of four change-point detection methods, namely, sequential probability ratio test (SPRT), cumulative sum (CUSUM), binomial event discriminator (BED), and online Bayesian change-point detection (OBCPD), by using residuals obtained from four water quality time series, chlorine, conductivity, total organic carbon, and turbidity. Our fundamental criterion for the performance evaluation of the four change-point detection methods is given by the receiver operating characteristic (ROC) curve which is characterized by the true positive rate as a function of the false positive rate. We highlight with detailed experiments that OBCPD provides the best performance for large contamination events, and we also provide insight on the choice of change-point detection algorithms to consider for designing efficient contamination detection schemes.

Capturing correlation changes by applying kernel change point detection on the running correlations

Modern industrial systems are becoming increasingly complex, posing severe challenges to traditional fault diagnosis methods. Traditional multivariate statistical analysis methods struggle with dynamic non-stationary data, incipient faults, and unknown fault types due to their stationarity assumptions and dependence on historical fault data. Existing research lacks systematic comparison between change point detection and traditional multivariate analysis methods, particularly regarding temporal dynamics capture and early warning capabilities. This study aims to compare the application characteristics and scenarios of change point detection versus traditional multivariate analysis in industrial fault diagnosis. Through systematic analysis of both methods' principles and characteristics, this work reveals that change point detection demonstrates superior performance in temporal dynamics capture, early warning, adaptability, and data efficiency, providing theoretical foundation for method selection and elucidating future directions for intelligent fault diagnosis in complex industrial environments.

Travel time parameters obtained from road traffic sensors data play an important role in traffic management practice. A travel time forecasting model is proposed for urban road traffic sensors data based on the method of change-point detection in this paper. The first-order differential operation is used for preprocessing over the actual loop data; a change-point detection algorithm is designed to classify the sequence of large number of travel time data items into several patterns; then a travel time forecasting model is established based on autoregressive integrated moving average (ARIMA) model. By computer simulation, different control parameters are chosen for adaptive change point search for travel time series, which is divided into several sections of similar state.Then linear weight function is used to fit travel time sequence and to forecast travel time. The results show that the model has high accuracy in travel time forecasting.

When the deep learning model is used to predict time series data, it is easy to be adversarially attacked. The time series data is sensitive to the abnormal disturbance and has strict requirements on the disturbance amount. To solve these problems, we propose to generate adversarial time series by adding disturbance terms to the original time series, and design an adversarial attack algorithm based on the importance measure (AAIM in short), which slightly perturbs the original data to improve the performance of the time series prediction model. The meteorological data of Guangzhou from 1980 to 2020 are selected as the studied time series data, which verifies that the proposed adversarial attack method not only effectively fools the target time series prediction model LSTNet, but also attacks the most advanced models based on CNN and RNN. It can be seen from the results of AAIM that the adversarial disturbance based on the first 5% importance measurement has almost the same effect as the 100% disturbance of the original time series. For the proposed AAIM, even if the disturbance is tiny, it can impact the prediction performance of the deep learning model significantly. Experimental results show that the proposed method achieves good transferability.

Change-point of multiple biomarkers in women with ovarian cancer

When the time dimension is added to datasets, time series data are obtained. Extracting knowledge from time series data requires special attention to the timing aspects of the data. An interesting activity in the field of knowledge discovery from time series data is predicting the timing of upcoming events. In this paper we present a method for mining fuzzy knowledge from time series data. In contrast to traditional time series analysis methods which largely focus on global models, our method is about the discovery of local patterns in time series. The extracted knowledge will be in the form of fuzzy association rules and it aims at predicting the approximate timing of upcoming events. The proposed method includes cleaning and filtering of time series data, segmenting time series, extracting important features for prediction, further cleaning on feature values, fuzzifying feature values, extracting fuzzy association rules, and pruning the discovered rules. We will show the efficiency of our approach on a stock market dataset.Keywordsknowledge discoverytime seriesfuzzy association Rulespreprocessingpost-processing

The increased availability of time series datasets prompts the development of new tools and methods that allow machine learning classifiers to better cope with time series data. Time series data are usually characterized by a high space dimensionality and a very strong correlation among features. This special nature makes the development of effective time series classifiers a challenging task. This work proposes and analyzes methods combining spectral decomposition and feature selection for time series classification problems and compares them against methods that work with original time series and time-dependent features. Briefly, our approach first applies discrete cosine transform (DCT) or discrete wavelet transform (DWT) on time series data. Then, it performs supervised feature selection/reduction by selecting only the most discriminative set of coefficients to represent the data. Experimental evaluations, carried out on multiple datasets, demonstrate the benefits of our approach in learning efficient and accurate time series classifiers.

This work presents the statistical analysis of time series of monthly average temperatures in several European locations using a state space approach, where it is considered a model with a deterministic seasonal component and a stochastic trend. Temperature rise rates in Europe seem to have increased in the last decades when compared with longer periods, hence change point detection methods were applied to residuals state space models in order to identify these possible changes in the monthly temperature rise rates. In Northern Europe the change points were, almost all, identified in the late 1980s while in Central and Southeastern Europe was, for the majority of cities, in the 1990s and later.