Search Results (24)

Time series data are prevalent in diverse fields such as manufacturing and sensor-based human activity recognition. In real-world applications, these data are often collected with variable sample lengths, which can pose challenges for classification models that typically require fixed-length inputs. Existing approaches either employ models designed to handle variable input sizes or standardize sample lengths before applying models; however, we contend that these approaches may compromise data integrity and ultimately reduce model performance. To address this issue, we propose Time series Into Pixels (TIP), an intuitive yet strong method that maps each time series data point into a pixel in 2D representation, where the vertical axis represents time steps and the horizontal axis captures the value at each timestamp. To evaluate our representation without relying on a powerful vision model as a backbone, we employ a straightforward LeNet-like 2D CNN model. Through extensive evaluations against 10 baseline models across 11 real-world benchmarks, TIP achieves 2–5% higher accuracy and 10–25% higher macro average precision. We also demonstrate that TIP performs comparably on complex multivariate data, with ablation studies underscoring the potential hazard of length normalization techniques in variable-length scenarios. We believe this method provides a significant advancement for handling variable-length time series data in real-world applications. The code is publicly available. Full article

To achieve automatic recognition of lightning images, which cannot easily be handled using the existing methods and still requires significant human resources, we propose a lightning image dataset and a preprocessing method. The lightning image data over five months were collected using a camera based on two optical observation stations, and then a series of batch labeling methods were applied, which greatly reduced the workload of subsequent manual labeling, and a dataset containing more than 30,000 labeled samples was established. Considering that lightning varies rapidly over time, we propose a time sequence composite (TSC) preprocessing method that inputs lightning’s time-varying characteristics into a model for better recognition of lightning images. The TSC method was evaluated through an experiment on four backbones, and it was found that this preprocessing method enhances the classification performance by 40%. The final trained model could successfully distinguish between the “lightning” and “non-lightning” samples, and a recall rate of 86.5% and a false detection rate of 0.2% were achieved. Full article

While deep neural networks (DNNs) have been widely and successfully used for time series classification (TSC) over the past decade, their vulnerability to adversarial attacks has received little attention. Most existing attack methods focus on white-box setups, which are unrealistic as attackers typically only have access to the model’s probability outputs. Defensive methods also have limitations, relying primarily on adversarial retraining which degrades classification accuracy and requires excessive training time. On top of that, we propose two new approaches in this paper: (1) A simulated annealing-based random search attack that finds adversarial examples without gradient estimation, searching only on the $l_{\infty }$-norm hypersphere of allowable perturbations. (2) A post-processing defense technique that periodically reverses the trend of corresponding loss values while maintaining the overall trend, using only the classifier’s confidence scores as input. Experiments applying these methods to InceptionNet models trained on the UCR dataset benchmarks demonstrate the effectiveness of the attack, achieving up to 100% success rates. The defense method provided protection against up to 91.24% of attacks while preserving prediction quality. Overall, this work addresses important gaps in adversarial TSC by introducing novel black-box attack and lightweight defense techniques. Full article

Time-Series-Classification (TSC) is an important machine learning task for many branches of science. Symbolic representations of time series, especially Symbolic Fourier Approximation (SFA), have been proven very effective for this task, given their abilities to reduce noise. In this paper, we improve upon SCALE-BOSS using multiple symbolic representations of time series. More specifically, the proposed SCALE-BOSS-MR incorporates into the process a variety of window sizes combined with multiple dilation parameters applied to the original and to first-order differences’ time series, with the latter modeling trend information. SCALE-BOSS-MR has been evaluated using the eight datasets with the largest training size of the UCR time series repository. The results indicate that SCALE-BOSS-MR can be instantiated to classifiers that are able to achieve state-of-the-art accuracy and can be tuned for scalability. Full article

Time series classification (TSC) is always a very important research topic in many real-world application domains. MultiRocket has been shown to be an efficient approach for TSC, by adding multiple pooling operators and a first-order difference transformation. To classify time series with higher accuracy, this study proposes a hybrid ensemble learning algorithm combining Complementary Ensemble Empirical Mode Decomposition (CEEMD) with improved MultiRocket, namely CEEMD-MultiRocket. Firstly, we utilize the decomposition method CEEMD to decompose raw time series into three sub-series: two Intrinsic Mode Functions (IMFs) and one residue. Then, the selection of these decomposed sub-series is executed on the known training set by comparing the classification accuracy of each IMF with that of raw time series using a given threshold. Finally, we optimize convolution kernels and pooling operators, and apply our improved MultiRocket to the raw time series, the selected decomposed sub-series and the first-order difference of the raw time series to generate the final classification results. Experiments were conducted on 109 datasets from the UCR time series repository to assess the classification performance of our CEEMD-MultiRocket. The extensive experimental results demonstrate that our CEEMD-MultiRocket has the second-best average rank on classification accuracy against a spread of the state-of-the-art (SOTA) TSC models. Specifically, CEEMD-MultiRocket is significantly more accurate than MultiRocket even though it requires a relatively long time, and is competitive with the currently most accurate model, HIVE-COTE 2.0, only with 1.4% of the computing load of the latter. Full article

Time-series data is an appealing study topic in data mining and has a broad range of applications. Many approaches have been employed to handle time series classification (TSC) challenges with promising results, among which deep neural network methods have become mainstream. Echo State Networks (ESN) and Convolutional Neural Networks (CNN) are commonly utilized as deep neural network methods in TSC research. However, ESN and CNN can only extract local dependencies relations of time series, resulting in long-term temporal data dependence needing to be more challenging to capture. As a result, an encoder and decoder architecture named LA-ESN is proposed for TSC tasks. In LA-ESN, the encoder is composed of ESN, which is utilized to obtain the time series matrix representation. Meanwhile, the decoder consists of a one-dimensional CNN (1D CNN), a Long Short-Term Memory network (LSTM) and an Attention Mechanism (AM), which can extract local information and global dependencies from the representation. Finally, many comparative experimental studies were conducted on 128 univariate datasets from different domains, and three evaluation metrics including classification accuracy, mean error and mean rank were exploited to evaluate the performance. In comparison to other approaches, LA-ESN produced good results. Full article

One possible device authentication method is based on device fingerprints, such as software- or hardware-based unique characteristics. In this paper, we propose a fingerprinting technique based on passive externally measured information, i.e., current consumption from the electrical network. The key insight is that small hardware discrepancies naturally exist even between same-electrical-circuit devices, making it feasible to identify slight variations in the consumed current under steady-state conditions. An experimental database of current consumption signals of two similar groups containing 20 same-model computer displays was collected. The resulting signals were classified using various state-of-the-art time-series classification (TSC) methods. We successfully identified 40 similar (same-model) electrical devices with about 94% precision, while most errors were concentrated in confusion between a small number of devices. A simplified empirical wavelet transform (EWT) paired with a linear discriminant analysis (LDA) classifier was shown to be the recommended classification method. Full article

Background: Digital clinical measures collected via various digital sensing technologies such as smartphones, smartwatches, wearables, and ingestible and implantable sensors are increasingly used by individuals and clinicians to capture the health outcomes or behavioral and physiological characteristics of individuals. Time series classification (TSC) is very commonly used for modeling digital clinical measures. While deep learning models for TSC are very common and powerful, there exist some fundamental challenges. This review presents the non-deep learning models that are commonly used for time series classification in biomedical applications that can achieve high performance. Objective: We performed a systematic review to characterize the techniques that are used in time series classification of digital clinical measures throughout all the stages of data processing and model building. Methods: We conducted a literature search on PubMed, as well as the Institute of Electrical and Electronics Engineers (IEEE), Web of Science, and SCOPUS databases using a range of search terms to retrieve peer-reviewed articles that report on the academic research about digital clinical measures from a five-year period between June 2016 and June 2021. We identified and categorized the research studies based on the types of classification algorithms and sensor input types. Results: We found 452 papers in total from four different databases: PubMed, IEEE, Web of Science Database, and SCOPUS. After removing duplicates and irrelevant papers, 135 articles remained for detailed review and data extraction. Among these, engineered features using time series methods that were subsequently fed into widely used machine learning classifiers were the most commonly used technique, and also most frequently achieved the best performance metrics (77 out of 135 articles). Statistical modeling (24 out of 135 articles) algorithms were the second most common and also the second-best classification technique. Conclusions: In this review paper, summaries of the time series classification models and interpretation methods for biomedical applications are summarized and categorized. While high time series classification performance has been achieved in digital clinical, physiological, or biomedical measures, no standard benchmark datasets, modeling methods, or reporting methodology exist. There is no single widely used method for time series model development or feature interpretation, however many different methods have proven successful. Full article

Early time series classification (ETSC) is crucial for real-world time-sensitive applications. This task aims to classify time series data with least timestamps at the desired accuracy. Early methods used fixed-length time series to train the deep models, and then quit the classification process by setting specific exiting rules. However, these methods may not adapt to the length variation of flow data in ETSC. Recent advances have proposed end-to-end frameworks, which leveraged the Recurrent Neural Networks to handle the varied-length problems, and the exiting subnets for early quitting. Unfortunately, the conflict between the classification and early exiting objectives is not fully considered. To handle these problems, we decouple the ETSC task into the varied-length TSC task and the early exiting task. First, to enhance the adaptive capacity of classification subnets to the data length variation, a feature augmentation module based on random length truncation is proposed. Then, to handle the conflict between classification and early exiting, the gradients of these two tasks are projected into a unified direction. Experimental results on 12 public datasets demonstrate the promising performance of our proposed method. Full article

Deep neural networks (DNN) have proven to be efficient in computer vision and data classification with an increasing number of successful applications. Time series classification (TSC) has been one of the challenging problems in data mining in the last decade, and significant research has been proposed with various solutions, including algorithm-based approaches as well as machine and deep learning approaches. This paper focuses on combining the two well-known deep learning techniques, namely the Inception module and the Fully Convolutional Network. The proposed method proved to be more efficient than the previous state-of-the-art InceptionTime method. We tested our model on the univariate TSC benchmark (the UCR/UEA archive), which includes 85 time-series datasets, and proved that our network outperforms the InceptionTime in terms of the training time and overall accuracy on the UCR archive. Full article

Time series classification (TSC) task is one of the most significant topics in data mining. Among all methods for this issue, the deep-learning-based shows superior performance for its good adaption to raw series data and automatic extraction of features. However, rare eyes are kept on composing ensembles of these superior individual classifiers to achieve further breakthroughs. The existing deep learning ensembles NNE did a heavy work of combining 60 individuals but did not maximize the deserving improvement, since it merely pays attention to the diversity of individuals but ignores their accuracy. In this paper, we propose to construct an ensemble of Full Convolutional Neural Networks (FCN) by Random Subspace Method (RSM), named RSM-FCN. FCN is a simple but outstanding individual classifier and RSM is suitable for high dimensional data such as time series, but there are few instances. Thus, the combination of these strengths, RSM-FCN provides a highly cost-effective approach to yield promising results. Experiments on the UCR dataset demonstrate the effectiveness and reasonability of the proposed method. Full article

Traditional supervised time series classification (TSC) tasks assume that all training data are labeled. However, in practice, manually labelling all unlabeled data could be very time-consuming and often requires the participation of skilled domain experts. In this paper, we concern with the positive unlabeled time series classification problem (PUTSC), which refers to automatically labelling the large unlabeled set U based on a small positive labeled set PL. The self-training (ST) is the most widely used method for solving the PUTSC problem and has attracted increased attention due to its simplicity and effectiveness. The existing ST methods simply employ the one-nearest-neighbor (1NN) formula to determine which unlabeled time-series should be labeled. Nevertheless, we note that the 1NN formula might not be optimal for PUTSC tasks because it may be sensitive to the initial labeled data located near the boundary between the positive and negative classes. To overcome this issue, in this paper we propose an exploratory methodology called ST-average. Unlike conventional ST-based approaches, ST-average utilizes the average sequence calculated by DTW barycenter averaging technique to label the data. Compared with any individuals in PL set, the average sequence is more representative. Our proposal is insensitive to the initial labeled data and is more reliable than existing ST-based methods. Besides, we demonstrate that ST-average can naturally be implemented along with many existing techniques used in original ST. Experimental results on public datasets show that ST-average performs better than related popular methods. Full article

Data analysis plays an increasingly valuable role in sports. The better the data that is analysed, the more concise training methods that can be chosen. Several solutions already exist for this purpose in the tennis industry; however, none of them combine data generation with a wristband and classification with a deep convolutional neural network (CNN). In this article, we demonstrate the development of a reliable shot detection trigger and a deep neural network that classifies tennis shots into three and five shot types. We generate a dataset for the training of neural networks with the help of a sensor wristband, which recorded 11 signals, including an inertial measurement unit (IMU). The final dataset included 5682 labelled shots of 16 players of age 13–70 years, predominantly at an amateur level. Two state-of-the-art architectures for time series classification (TSC) are compared, namely a fully convolutional network (FCN) and a residual network (ResNet). Recent advances in the field of machine learning, like the Mish activation function and the Ranger optimizer, are utilized. Training with the rather inhomogeneous dataset led to an F₁ score of 96% in classification of the main shots and 94% for the expansion. Consequently, the study yielded a solid base for more complex tennis analysis tools, such as the indication of success rates per shot type. Full article

Time series classification (TSC) is a significant problem in data mining with several applications in different domains. Mining different distinguishing features is the primary method. One promising method is algorithms based on the morphological structure of time series, which are interpretable and accurate. However, existing structural feature-based algorithms, such as time series forest (TSF) and shapelet traverse, all features through many random combinations, which means that a lot of training time and computing resources are required to filter meaningless features, important distinguishing information will be ignored. To overcome this problem, in this paper, we propose a perceptual features-based framework for TSC. We are inspired by how humans observe time series and realize that there are usually only a few essential points that need to be remembered for a time series. Although the complex time series has a lot of details, a small number of data points is enough to describe the shape of the entire sample. First, we use the improved perceptually important points (PIPs) to extract key points and use them as the basis for time series segmentation to obtain a combination of interval-level and point-level features. Secondly, we propose a framework to explore the effects of perceptual structural features combined with decision trees (DT), random forests (RF), and gradient boosting decision trees (GBDT) on TSC. The experimental results on the UCR datasets show that our work has achieved leading accuracy, which is instructive for follow-up research. Full article

Recently, some researchers adopted the convolutional neural network (CNN) for time series classification (TSC) and have achieved better performance than most hand-crafted methods in the University of California, Riverside (UCR) archive. The secret to the success of the CNN is weight sharing, which is robust to the global translation of the time series. However, global translation invariance is not the only case considered for TSC. Temporal distortion is another common phenomenon besides global translation in time series. The scale and phase changes due to temporal distortion bring significant challenges to TSC, which is out of the scope of conventional CNNs. In this paper, a CNN architecture with an elastic matching mechanism, which is named Elastic Matching CNN (short for EM-CNN), is proposed to address this challenge. Compared with the conventional CNN, EM-CNN allows local time shifting between the time series and convolutional kernels, and a matching matrix is exploited to learn the nonlinear alignment between time series and convolutional kernels of the CNN. Several EM-CNN models are proposed in this paper based on diverse CNN models. The results for 85 UCR datasets demonstrate that the elastic matching mechanism effectively improves CNN performance. Full article