Figure 1 - uploaded by Qingsong Wen
Content may be subject to copyright.
A taxonomy of time series data augmentation techniques.
Source publication
Deep learning performs remarkably well on many time series analysis tasks recently. The superior performance of deep neural networks relies heavily on a large number of training data to avoid overfitting. However, the labeled data of many real-world time series applications may be limited such as classification in medical time series and anomaly de...
Contexts in source publication
Context 1
... paper, we aim to fill the aforementioned gaps by summarizing existing time series data augmentation methods in common tasks, including time series forecasting, anomaly detection, classification, as well as providing insightful future directions. To this end, we propose a taxonomy of data augmentation methods for time series, as illustrated in Fig. 1. Based on the taxonomy, we review these data augmentation methods systematically. We start the discussion from the sim- ple transformations in time domain first. And then we discuss more transformations on time series in the transformed frequency and time-frequency domains. Besides the transformations in different domains for time ...
Context 2
... different data augmentation methods summarized in Fig. 1, one key strategy is how to select and combine various augmentation methods together. The experiments in [Um et al., 2017] show that the combination of three basic timedomain methods (permutation, rotation, and time warping) is better than that of a single method and achieves the best performance in time series classification. Also, ...
Context 3
... paper, we aim to fill the aforementioned gaps by summarizing existing time series data augmentation methods in common tasks, including time series forecasting, anomaly detection, classification, as well as providing insightful future directions. To this end, we propose a taxonomy of data augmentation methods for time series, as illustrated in Fig. 1. Based on the taxonomy, we review these data augmentation methods systematically. We start the discussion from the sim- ple transformations in time domain first. And then we discuss more transformations on time series in the transformed frequency and time-frequency domains. Besides the transformations in different domains for time ...
Context 4
... different data augmentation methods summarized in Fig. 1, one key strategy is how to select and combine various augmentation methods together. The experiments in [Um et al., 2017] show that the combination of three basic timedomain methods (permutation, rotation, and time warping) is better than that of a single method and achieves the best performance in time series classification. Also, ...
Citations
... According to this view, the main drawback of a full Bayesian estimate is its prohibitive cost, which leads to a very active search for approximations that offer the best trade-off between accuracy and computational efficiency (Blundell et al., 2015;Gal & Ghahramani, 2016;Hartmann & Richter, 2023;Jospin et al., 2020;MacKay, 1992;Sensoy et al., 2018;Titterington, 2004). Besides the Bayesian framework, the other main approaches rely either on ensemble methods (Lakshminarayanan et al., 2017;Michelucci & Venturini, 2021;Tavazza et al., 2021;Wen et al., 2020), or data augmentation methods (Shorten & Khoshgoftaar, 2019;Wen et al., 2021). An alternative is to train the DNN to specifically identify outliers or uncertain predictions [see e.g. ...
In recent years, the question of the reliability of Machine Learning (ML) methods has acquired significant importance, and the analysis of the associated uncertainties has motivated a growing amount of research. However, most of these studies have applied standard error analysis to ML models—and in particular Deep Neural Network (DNN) models—which represent a rather significant departure from standard scientific modelling. It is therefore necessary to integrate the standard error analysis with a deeper epistemological analysis of the possible differences between DNN models and standard scientific modelling and the possible implications of these differences in the assessment of reliability. This article offers several contributions. First, it emphasises the ubiquitous role of model assumptions (both in ML and traditional science) against the illusion of theory-free science. Secondly, model assumptions are analysed from the point of view of their (epistemic) complexity, which is shown to be language-independent. It is argued that the high epistemic complexity of DNN models hinders the estimate of their reliability and also their prospect of long term progress. Some potential ways forward are suggested. Thirdly, this article identifies the close relation between a model’s epistemic complexity and its interpretability, as introduced in the context of responsible AI. This clarifies in which sense—and to what extent—the lack of understanding of a model (black-box problem) impacts its interpretability in a way that is independent of individual skills. It also clarifies how interpretability is a precondition for a plausible assessment of the reliability of any model, which cannot be based on statistical analysis alone. This article focuses on the comparison between traditional scientific models and DNN models. However, Random Forest (RF) and Logistic Regression (LR) models are also briefly considered.
... For time series, there are various methods to create synthetic time series. These range from simple methods such as jittering or flipping the original time series [31], to more sophisticated approaches such as seasonal decomposition [12] or generative models [20]. ...
The effectiveness of univariate forecasting models is often hampered by conditions that cause them stress. A model is considered to be under stress if it shows a negative behaviour, such as higher-than-usual errors or increased uncertainty. Understanding the factors that cause stress to forecasting models is important to improve their reliability, transparency, and utility. This paper addresses this problem by contributing with a novel framework called MAST (Meta-learning and data Augmentation for Stress Testing). The proposed approach aims to model and characterize stress in univariate time series forecasting models, focusing on conditions where they exhibit large errors. In particular, MAST is a meta-learning approach that predicts the probability that a given model will perform poorly on a given time series based on a set of statistical time series features. MAST also encompasses a novel data augmentation technique based on oversampling to improve the metadata concerning stress. We conducted experiments using three benchmark datasets that contain a total of 49.794 time series to validate the performance of MAST. The results suggest that the proposed approach is able to identify conditions that lead to large errors. The method and experiments are publicly available in a repository.
... Data augmentation can effectively expand the data samples and prevent overfitting of the training model, which is critical for the successful use of deep learning models as it is a useful tool for increasing the quality and dimension of the input features [43]. It has been shown to be effective in many applications such as time series forecasting [44], [45]. In addition, Data augmentation can minimize sensor inputs to reduce the requirements of marker data and sensor for the estimation of gait variables [46], thus have potential for gait analysis as well as assistive device design. ...
Many challenges exist in the study of using orthotics, exoskeletons or exosuits as tools for rehabilitation and assistance of healthy people in daily activities due to the requirements of portability and safe interaction with the user and the environment. One approach to dealing with these challenges is to design a control system that can be deployed in a portable device to identify the relationships that exist between the gait variables and gait cycle for different locomotion modes. In order to estimate the knee and ankle angles in the sagittal plane for different locomotion modes, a novel multimodal feature-decoupled kinematic estimation system consisting of a multimodal locomotion classifier and an optimal joint angle estimator is proposed in this paper. The multi-source information output from different conventional primary models are fused by assigning the non-fixed weight. To improve the performance of the primary models, a data augmentation module based on the time-frequency domain analysis method is designed. The results show that the inclusion of the data augmentation module and multi-source information fusion modules has improved the classification accuracy to 98.56% and kinematic estimation performance (PCC) to 0.904 (walking), 0.956 (running), 0.899 (stair ascent), 0.851 (stair descent), respectively. The kinematic estimation quality is generally higher for faster speed (running) or proximal joint (knee) compared to other modes and ankle. The limitations and advantages of the proposed approach are discussed. Based on our findings, the multimodal kinematic estimation system has potential in facilitating the deployment for human-in-loop control of lower-limb intelligent assistive devices.
... For time series augmentation, a diverse set of techniques was proposed like adding jitter, scaling the data up or down, or adding shifts. For surveys grouping these techniques, see for example Iglesias et al. (2023), Iwana and Uchida (2021) or Wen et al. (2021). Compared to ML studies proposing new time series augmentation techniques, we focus on a robustness evaluation of time shifts gathered from field data on the prediction performance. ...
High quality data is essential to the success of machine learning projects, especially for training, but also after deployment. Even slight differences between training and runtime data may degrade performance. Based on the application case of truck driver stress prediction, we collected physiological, activity, and driving data using an Apple Watch 7, heart rate data using an ECG and weather data from a web service. We experimentally evaluated the prediction performance of increasing time-shifts applied to our data sources. Such problems are known as Out-of-Distribution situations. In this paper, we showcase how developers can approach such problems and perform analyses to identify features highly prone to Out-of-Distribution issues. These results are central to quality assurance for successful Machine Learning projects. We also propose Data Robustness Stories to document Out-of-Distribution issues.
... This can cause overfitting and lower the performance of models. To address this, the study employs time-series data augmentation techniques such as Time Warping, Noise Injection, Smoothing, and Trend Shifting to alleviate class imbalance and prevent overfitting while processing the data with deep learning models [9], [10], [11], [12], [13], [14]. Each augmentation method's analysis and performance evaluation reveal that noise injection is the most effective. ...
... LSTM (Long Short-Term Memory) [14] is a model developed to address the issue of long-term dependencies, a limitation inherent in traditional RNNs (Recurrent Neural Networks) [24]. Fig. 4 presents the process of the LSTM model used for multi-class classification performance evaluation. ...
... Table 2 shows the performance evaluation of noise injection augmentation techniques based on LSTM, GRU, and TCN models. According to the performance evaluation results in Table 2, the LSTM (Long Short-Term Memory) [14] model exhibits higher F1-Score, Precision, and Recall compared to the GRU (Gated Recurrent Unit) [15] and TCN (Temporal Convolutional Network) [16] models. Compared to other models, the LSTM model has a complex structure and many parameters. ...
This paper proposes a new approach to classify and evaluate defects in concrete structures automatically. To overcome the limitations of defect detection methods that traditionally relied on expert visual observation, the reflection signal of electromagnetic pulses is extracted as time-series data and used to analyze the propagation characteristics of each defect. This study uses deep learning models to analyze these time-series data and classify defects. Since anomaly detection data has more normal data than anomaly data, data augmentation methods such as Time Warping, Noise Injection, Smoothing, Trend Shifting, etc., were applied to solve the problem of data imbalance and overfitting. Among them, Noise Injection showed the best performance. The generalization performance of the proposed method was evaluated through performance evaluation using LSTM, GRU, and TCN models, and LSTM models showed the highest performance. The study results show that the proposed method effectively classifies defect types in concrete structures and can solve the limitations of existing methods by automatic classification through deep learning models. In addition, it was confirmed that the model's performance could be improved by improving the amount and diversity of data by selecting and applying appropriate data augmentation methods. The contribution of the research is to present a new approach that automates the defect detection and classification task of concrete structures and provides high accuracy and efficiency.
... Most of the early cutting-edge Convolutional Neural Network (CNN) [19] architectures used data augmentation, such as cropping [20], scaling [21], mirroring [22] and colour augmentation on images [23][24][25]. Although data augmentation is frequently used in neural network-based image identification, it is not a recognised best practice for time series recognition [26]. Compared to data augmentation for images, stochastic transformations of the training data for the time series data have not been explored thoroughly. ...
Recently, the research community has shown significant interest in the continuous temporal data obtained from motion sensors in wearable devices. These data are useful for classifying and analysing different human activities in many application areas such as healthcare, sports and surveillance. The literature has presented a multitude of deep learning models that aim to derive a suitable feature representation from temporal sensory input. However, the presence of a substantial quantity of annotated training data is crucial to adequately train the deep networks. Nevertheless, the data originating from the wearable devices are vast but ineffective due to a lack of labels which hinders our ability to train the models with optimal efficiency. This phenomenon leads to the model experiencing overfitting. The contribution of the proposed research is twofold: firstly, it involves a systematic evaluation of fifteen different augmentation strategies to solve the inadequacy problem of labeled data which plays a critical role in the classification tasks. Secondly, it introduces an automatic feature-learning technique proposing a Multi-Branch Hybrid Conv-LSTM network to classify human activities of daily living using multimodal data of different wearable smart devices. The objective of this study is to introduce an ensemble deep model that effectively captures intricate patterns and interdependencies within temporal data. The term “ensemble model” pertains to fusion of distinct deep models, with the objective of leveraging their own strengths and capabilities to develop a solution that is more robust and efficient. A comprehensive assessment of ensemble models is conducted using data-augmentation techniques on two prominent benchmark datasets: CogAge and UniMiB-SHAR. The proposed network employs a range of data-augmentation methods to improve the accuracy of atomic and composite activities. This results in a 5% increase in accuracy for composite activities and a 30% increase for atomic activities.
... By artificially enhancing dataset size and diversity, data augmentation techniques have proven to significantly mitigate overfitting, thereby improving model robustness and performance [3]. This success has sparked interest in applying similar strategies within the domain of time series classification, where the challenges of data scarcity and class imbalance are equally prevalent [6,7,8,9]. ...
... This makes them a promising candidate for multivariate time series analysis, and we evaluate them in our work. Our methodology encompasses a diverse range of augmentation strategies, each carefully selected to enhance the representativeness and quality of the training data, thereby enabling models to achieve superior generalization and performance [6]. ...
... Our taxonomy sets itself apart from other taxonomies [7,6] by incorporating the preserving class of techniques, which try to address the following challenges. First, when performing data augmentation by adding noise, how can we determine the optimal amount of noise to augment a series intelligently? ...
Our study investigates the impact of data augmentation on the performance of multivariate time series models, focusing on datasets from the UCR archive. Despite the limited size of these datasets, we achieved classification accuracy improvements in 10 out of 13 datasets using the Rocket and InceptionTime models. This highlights the essential role of sufficient data in training effective models, paralleling the advancements seen in computer vision. Our work delves into adapting and applying existing methods in innovative ways to the domain of multivariate time series classification. Our comprehensive exploration of these techniques sets a new standard for addressing data scarcity in time series analysis, emphasizing that diverse augmentation strategies are crucial for unlocking the potential of both traditional and deep learning models. Moreover, by meticulously analyzing and applying a variety of augmentation techniques, we demonstrate that strategic data enrichment can enhance model accuracy. This not only establishes a benchmark for future research in time series analysis but also underscores the importance of adopting varied augmentation approaches to improve model performance in the face of limited data availability.
... Such models demand substantial computational power, thereby leading to elevated research costs. Consequently, it is crucial to integrate pertinent key technologies for lightweighting GAIM, including reinforcement learning from human feedback (RLHF) [71], prompt learning [72], knowledge graph [73], and related lightweight techniques like pruning [74], distilling [75], and data augmentation [76]. The integration aims to reduce the training parameters while preserving the model's functionality, ultimately lowering the training cost. ...
Process planning serves as a critical link between design and manufacturing, exerting a pivotal influence on the quality and efficiency of production. However, current intelligent process planning systems, like computer-aided process planning (CAPP), still contend with the challenge of realizing comprehensive automation in process decision-making. These obstacles chiefly involve, though are not confined to, issues like limited intelligence, poor flexibility, low reliability, and high usage thresholds. Generative artificial intelligence (AI) has attained noteworthy accomplishments in natural language processing (NLP), offering new perspectives to address these challenges. This paper summarizes the limitations of current intelligent process planning methods and explores the potential of integrating generative AI into process planning. With synergistically incorporating digital twin (DT), this paper introduces a conceptual framework termed generative AI and DT-enabling intelligent process planning (GIPP). The paper elaborates on two supporting methodologies: process generative pre-trained transformer (GPT) modelling and DT-based process verification method. Moreover, a prototype system is established to introduce the implementation and machining execution mechanism of GIPP for milling a specific thin-walled component. Three potential application scenarios and a comparative analysis are employed to elucidate the practicality of GIPP, providing new insights for intelligent process planning.
... Recent advancements in machine learning techniques have shown promise in capturing such complex patterns in time series data with distributional shifts. However, the effectiveness of these learning-based methods is often limited by the quantity of data samples [7,8]. The data scarcity issue can be raised from various aspects. ...
Effective utilization of time series data is often constrained by the scarcity of data quantity that reflects complex dynamics, especially under the condition of distributional shifts. Existing datasets may not encompass the full range of statistical properties required for robust and comprehensive analysis. And privacy concerns can further limit their accessibility in domains such as finance and healthcare. This paper presents an approach that utilizes large language models and data source interfaces to explore and collect time series datasets. While obtained from external sources, the collected data share critical statistical properties with primary time series datasets, making it possible to model and adapt to various scenarios. This method enlarges the data quantity when the original data is limited or lacks essential properties. It suggests that collected datasets can effectively supplement existing datasets, especially involving changes in data distribution. We demonstrate the effectiveness of the collected datasets through practical examples and show how time series forecasting foundation models fine-tuned on these datasets achieve comparable performance to those models without fine-tuning.
... We calculate the mean of the outcomes to derive a final assessment. To rectify disparities in data classes, we also incorporated window slicing augmentation [57], which involves the selection of random, contiguous segments from the electronic health records of patients. ...
Sepsis is a severe and expensive medical emergency that requires prompt identification in order to improve patient mortality. The objective of our research is to develop an attention-based bidirectional LSTM-CNN (AT-BiLSTM-CNN) hybrid architecture for the early prediction of sepsis using electronic health records (EHRs) obtained from intensive care units (ICUs). We combine attention mechanism, bidirectional long short-term memory (BiLSTM) and convolutional neural network (CNN) to analyse clinical time series data, aiming to enhance prediction accuracy. The effectiveness of our model is measured using metrics such as accuracy, sensitivity, specificity, and area under the receiver operating characteristic (AUROC), utilising data from the 2019 PhysioNet Challenge. Upon assessing the performance of the AT-BiLSTM-CNN model throughout prediction windows of 4, 8, and 12 h, we observed its exceptional performance in comparison with existing leading techniques. It achieved average AUROCs of 0.88, 0.85, and 0.84 for the predictions made 4, 8, and 12 h before sepsis onset, respectively. This research contributes significantly to the development of smart clinical support systems, potentially offering lifesaving interventions for septic patients at critical moments.
