Article
Energy & Fuels
Marco G. Pinheiro, Sara C. Madeira, Alexandre P. Francisco
Summary: Data is an important asset for the electric power industry, supporting management decisions, operational efficiency, and competitiveness. The advent of smart grids has increased the availability of data, but the inability to recognize the value of data beyond its specific application is seen as a barrier. Power load time series data is crucial for utilities due to its inherent information on human behavior, economic trends, and other factors. Time series analysis in the energy sector has gained interest due to the growing availability of data from sensorization in power grids. This study demonstrates the effectiveness of the shapelet technique in creating interpretable classifiers for different hierarchical power levels, showing its potential for extracting interpretable patterns and knowledge and recognizing the value of data in driving services within the energy sector.
Article
Computer Science, Artificial Intelligence
Hussein El Amouri, Thomas Lampert, Pierre Gancarski, Clement Mallet
Summary: The analysis of time series is increasingly popular due to sensor proliferation. Dynamic Time Warping (DTW) is a widely used similarity measure for time series, but it violates metric axioms. Learning DTW-Preserving Shapelets (LDPS) reintroduces these axioms. This article extends LDPS to constrained DTW-preserving shapelets (CDPS) that consider user knowledge in the form of must-link and cannot-link constraints, allowing for clustering that respects the constraints.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Artificial Intelligence
Youssef Hmamouche, Lotfi Lakhal, Alain Casali
Summary: This article presents a scalable prediction process for large time series prediction, including a new algorithm for identifying time series predictors. The proposed framework shows promising results on real datasets and improves the prediction accuracy of many time series.
KNOWLEDGE AND INFORMATION SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Mubarak G. Abdu-Aguye, Walid Gomaa, Yasushi Makihara, Yasushi Yagi
Summary: This paper proposes axioms of robustness against adversarial attacks in time series classification and empirically validates the hypotheses put forth.
PATTERN RECOGNITION LETTERS
(2022)
Article
Statistics & Probability
Roberto Medico, Joeri Ruyssinck, Dirk Deschrijver, Tom Dhaene
Summary: Shapelets are discriminative subsequences extracted from time-series data for enhancing classifier performance and interpretability of the model. This research introduces a novel architecture and learning strategy for multi-dimensional Shapelets learning, automatically selecting smaller sets of uncorrelated Shapelets and achieving competitive performance across datasets.
ADVANCES IN DATA ANALYSIS AND CLASSIFICATION
(2021)
Article
Environmental Sciences
Maria Yli-Heikkila, Samantha Wittke, Markku Luotamo, Eetu Puttonen, Mika Sulkava, Petri Pellikka, Janne Heiskanen, Arto Klami
Summary: One of the key principles of food security is to ensure the proper functioning of global food markets. This study proposes a method for large-scale crop yield estimations using satellite image time series, and demonstrates that a deep learning-based temporal convolutional network outperforms traditional machine learning methods and national crop forecasts in accuracy. The study also shows that mean-aggregated regional predictions with histogram-based features calculated from farm-level observations perform better than other tested approaches.
Article
Computer Science, Artificial Intelligence
Jiahui Chen, Yuan Wan, Xiaoyu Wang, Yinglv Xuan
Summary: This paper proposes a novel learning-based shapelet discovery method for time series classification, which improves the accuracy and interpretability of shapelets through feature selection.
APPLIED INTELLIGENCE
(2022)
Article
Computer Science, Artificial Intelligence
Fazle Karim, Somshubra Majumdar, Houshang Darabi
Summary: This paper proposes a method to attack time series classification models using adversarial samples, demonstrating attacks on 42 datasets. The proposed attack generates a larger fraction of successful adversarial black-box attacks compared to the Fast Gradient Sign Method, and a simple defense mechanism is successfully devised to reduce the success rate of adversarial samples. Future researchers are recommended to incorporate adversarial data samples into their training datasets to enhance resilience against adversarial samples.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2021)
Article
Chemistry, Analytical
Kenan Li, Huiyu Deng, John Morrison, Rima Habre, Meredith Franklin, Yao-Yi Chiang, Katherine Sward, Frank D. Gilliland, Jose Luis Ambite, Sandrah P. Eckel
Summary: Time series classification often relies on machine learning, but there is growing interest in understanding discriminatory features of time series beyond black box models. Time-series shapelets (TSS) is a promising method for identifying discriminative subsequences, with the novel intelligent method Wavelet-TSS (W-TSS) using wavelet transformation discovery for candidate shapelet identification. Compared to previous TSS algorithms, W-TSS is more computationally efficient, accurate, and able to discover more discriminative shapelets without the need for pre-specification of shapelet length.
Article
Computer Science, Artificial Intelligence
Chang Wei Tan, Christoph Bergmeir, Francois Petitjean, Geoffrey I. Webb
Summary: This paper studies time series extrinsic regression (TSER) and finds that the Rocket TSC algorithm achieves the highest accuracy when adapted for regression. More research is needed to improve the accuracy of ML models in this field, with good prospects for further advancements beyond straightforward baselines.
DATA MINING AND KNOWLEDGE DISCOVERY
(2021)
Article
Computer Science, Artificial Intelligence
S. Abilasha, Sahely Bhadra, P. Deepak, Anish Mathew
Summary: Time series data is widely used in various domains, and detecting anomalies in such data is crucial. However, traditional methods often fail to handle time series with warping variations effectively. In this paper, a novel anomaly detection method is proposed that utilizes data augmentation and a twin autoencoder architecture to learn warping-robust representations for time series data, achieving improved anomaly detection performance.
Article
Engineering, Civil
Georgia Papacharalampous, Hristos Tyralis, Yannis Markonis, Martin Hanel
Summary: In this study, a new methodological framework is proposed for exploring and comparing multi-scale analyses in a hydroclimatic context, in order to comprehensively understand the behaviors of geophysical processes and evaluate time series simulation models. By computing the feature values at different temporal resolutions and three hydroclimatic time series types, similarities and differences in the evolution patterns are identified. The computed features are also used for meaningful clustering of hydroclimatic time series, which allows for interpretation of hydroclimatic similarity at various temporal resolutions.
JOURNAL OF HYDROLOGY
(2023)
Article
Computer Science, Artificial Intelligence
Ali Jameel Hashim, M. A. Balafar, Jafar Tanha, Aryaz Baradarani
Summary: In recent years, anomaly detection has made significant advancements due to the increasing demand for identifying outliers in various engineering applications that experience environmental adaptations. Researchers have developed an adaptive evolutionary autoencoder approach for anomaly detection in time-series data, leveraging the integration of unsupervised machine learning techniques with evolutionary intelligence. The effectiveness, speed, and functionality enhancements of this method are demonstrated through its implementation, and a comprehensive statistical analysis validates its advantages in unsupervised anomaly detection.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Environmental Sciences
Laura Crocetti, Matthias Schartner, Benedikt Soja
Summary: This study investigates the use of machine learning algorithms to detect discontinuities caused by earthquakes in GNSS time series, and finds that Random Forest algorithm performs the best. Splitting the time series into chunks of 21 days, combining the components into one sample, and adding value range as an additional feature improve the detection results.
Article
Automation & Control Systems
Samuel Harford, Fazle Karim, Houshang Darabi
Summary: This study proposes a method for generating adversarial samples on multivariate time series classification models, combining adversarial autoencoders and gradient adversarial transformation networks. By utilizing adversarial attacks, the adversarial samples are improved by replacing the adversarial generator function with variational autoencoders.
IEEE-CAA JOURNAL OF AUTOMATICA SINICA
(2021)