Time Series Data Research Articles

COVID-19 health data prediction: a critical evaluation of CNN-based approaches

The COVID-19 pandemic has significantly accelerated the demand for accurate and efficient prediction models to support effective disease management, containment strategies, and informed decision-making. Predictive models capable of analyzing complex health data are essential for monitoring disease trends, evaluating risk factors, and optimizing resource allocation during the pandemic. Among various machine learning approaches, convolutional neural networks (CNNs) have emerged as powerful tools due to their ability to process large volumes of high-dimensional health data, such as medical images, time-series data, and patient demographics, with impressive precision. This research seeks to systematically examine the challenges and limitations inherent in utilizing CNNs for COVID-19 health data prediction, offering a comprehensive perspective grounded in data science research. Key areas of investigation include issues related to data quality and availability, such as incomplete, noisy, and imbalanced datasets, which often hinder the training of robust models. Additionally, architectural constraints of CNNs, including their sensitivity to hyperparameter tuning and reliance on substantial computational resources, are explored as critical bottlenecks that impact scalability and efficiency. A significant focus is placed on generalization challenges, where models trained on specific datasets struggle to adapt to unseen data from diverse populations or clinical settings, limiting their applicability in real-world scenarios. The study further highlights a reported accuracy of 63%, underscoring the need for improved methodologies to enhance model performance and reliability. By addressing these challenges, this research aims to provide actionable insights and practical recommendations to optimize the use of CNNs for COVID-19 health data prediction. In particular, the study emphasizes the importance of incorporating advanced strategies such as transfer learning, data augmentation, and regularization techniques to overcome dataset limitations and enhance model robustness. The integration of multimodal approaches combining medical images with auxiliary data, such as patient demographics and laboratory results, is proposed to improve contextual understanding and diagnostic precision. Finally, the research underscores the necessity of interdisciplinary collaboration, leveraging domain expertise from data scientists, healthcare professionals, and epidemiologists to develop holistic solutions for tackling the complexities of COVID-19 prediction. By shedding light on the limitations and potential of CNNs in this domain, this study aims to guide researchers and practitioners in making informed decisions about model design, implementation, and optimization. Ultimately, it contributes to advancing AI-driven diagnostics and predictive modeling for COVID-19 and other public health crises, fostering the development of scalable and reliable tools for better healthcare outcomes.

Economic growth and carbon emissions nexus: environmental sustainability a case of japan from East Asia

Economic growth plays a vital and instrumental role in the economic development of any country. Still, on the other side, it is also one of the important causes of environmental degradation. The increase in carbon emissions is one of the fundamental reasons for climate change and global warming. Japan’s global economic and technological leadership, combined with its ambitious carbon neutrality goals by 2050, makes it a compelling case for understanding the interplay between economic growth, renewable energy adoption, and carbon emissions. We took annual time series data for the dependent variable as Carbon Emissions and explanatory variables as Gross Domestic Product and Renewable Energy for the period ranging from 1990 to 2020. We applied the Autoregressive Distributed Lags Bounds test to gauge the long-run relationship between the variables; the empirical findings revealed the existence of a long-run relationship among the variables of the model. GDP has a significant positive impact on carbon emissions whereas renewable energy has a significant negative impact on carbon emissions. Furthermore, we also found renewable energy is unidirectional causing CO2. Based on these findings, we recommend targeted policies to reduce carbon emissions, including stricter regulations for high-emission industries and expanded investments in renewable energy infrastructure. These measures will support Japan's path to carbon neutrality while fostering sustainable economic growth.

Development of a cryptocurrency price prediction model: leveraging GRU and LSTM for Bitcoin, Litecoin and Ethereum

Cryptocurrency represents a form of asset that has arisen from the progress of financial technology, presenting significant prospects for scholarly investigations. The ability to anticipate cryptocurrency prices with extreme accuracy is very desirable to researchers and investors. However, time-series data presents significant challenges due to the nonlinear nature of the cryptocurrency market, complicating precise price predictions. Several studies have explored cryptocurrency price prediction using various deep learning (DL) algorithms. Three leading cryptocurrencies, determined by market capitalization, Ethereum (ETH), Bitcoin (BTC), and Litecoin (LTC), are examined for exchange rate predictions in this study. Two categories of recurrent neural networks (RNNs), specifically long short-term memory (LSTM) and gated recurrent unit (GRU), are employed. Four performance metrics are selected to evaluate the prediction accuracy namely mean squared error (MSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean squared error (RMSE) for three cryptocurrencies which demonstrates that GRU model outperforms LSTM. The GRU model was implemented as a two-layer deep learning network, optimized using the Adam optimizer with a dropout rate of 0.2 to prevent overfitting. The model was trained using normalized historical price data sourced from CryptoDataDownload, with an 80:20 train-test split. In this work, GRU qualifies as the best algorithm for developing a cryptocurrency price prediction model. MAPE values for BTC, LTC and ETH are 0.03540, 0.08703 and 0.04415, respectively, which indicate that GRU offers the most accurate forecasts as compared to LSTM. These prediction models are valuable for traders and investors, offering accurate cryptocurrency price predictions. Future studies should also consider additional variables, such as social media trends and trade volumes that may impact cryptocurrency pricing.

Selected Macroeconomic Indicators and Economic Wellbeing in Nigeria

This study examines selected macroeconomic indicators and economic well-being in Nigeria. It focuses on the effects of exchange rate fluctuation, inflation, balance of payments, unemployment and literacy rate in both short and long run on the well-being of Nigerians. It used econometric analysis on the annual time series data collected from both the Central Bank of Nigeria Statistical Bulletin and World Bank development indications of 2024. The findings of the study reveal that exchange rate is negative and statistically significant on the economic well-being in the long run indicating, that depreciation of the Naira reduces purchasing power and worsens economic conditions. Though inflation has a negative impact, is found to be statistically insignificant, suggesting the presence of economic adaptation mechanisms. Balance of payments was found to have positive and significant effect, demonstrating that an improved trade balance enhances economic well-being. However, unemployment and literacy rates, despite having positive coefficients, are found to be statistically insignificant. This implies that structural labor market inefficiencies and educational mismatch may limit their direct impact on economic well-being. The study concludes that macroeconomic stability, exchange rate management, trade balance improvements, and job creation strategies are essential for enhancing economic well-being in Nigeria. The study therefore made the following policy recommendations; ensuring foreign exchange stability measures, export diversification, inflation control policies, labor market reforms, and investments in infrastructure and education. Implementation of these strategies will contribute to sustainable economic growth and improved living standards for Nigerians. JEL: 011,023,047.

Highway non-recurrent congestion prediction using a multi-step spatio-temporal deep learning approach

ABSTRACT Accurate forecasting of highway Non-Recurrent Congestion (NRC) is critical for modern transportation systems. However, this research remains in its early stages and is frequently constrained to single-step temporal prediction. To address this limitation, this research presents a novel approach that leverages the Dual-Stream Autoencoder Sequence-to-Sequence model (DS-AE-Seq2Seq) to create an accurate tool for quantitative, spatio-temporal, and multi-step prediction of highway NRC. The proposed model innovatively integrates an Autoencoder and a Seq2Seq encoder to process static and time-series data, respectively. The decoder generates spatio-temporal predictions using the outcomes of both streams. The model is tested with data collected from highway I-5 and I-405, USA. Results show that it not only outperforms benchmarks but also exhibits high prediction accuracy under extreme traffic conditions, including severe injury incidents and various levels of service. Additionally, the model demonstrated reliability, through a sensitivity analysis, across different distances and prediction horizons.

Performance Evaluation of Long Short-Term Memory and Autoregressive Integrated Moving Average Time Series Models for Stock Price Prediction

Abstract: Stock price prediction is a critical task in financial markets, often complicated by the challenges of modeling highly volatile, non-linear, and dynamic time series data. This study evaluates the performance of two prominent forecasting models: Long Short-Term Memory (LSTM) networks, known for their ability to capture long-term dependencies and non-linear patterns, and the Auto-Regressive Integrated Moving Average (ARIMA), a traditional statistical model adept at linear trend modeling. Historical stock price data from the Nigerian Exchange Limited (NGX) was utilized. Both models were implemented in a Python environment, and their predictive accuracy was assessed using performance metrics, including Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and R-squared (R²). Additionally, the Diebold-Mariano (DM) test was employed to statistically compare the models’ predictive accuracies. This study highlights the potential of LSTM models for robust stock price forecasting and provides valuable insights for selecting predictive models based on data complexity and market dynamics.

Pengaruh Indeks Pembangunan Manusia, Pengangguran Dan Tenaga Kerja Terhadap Kemiskinan Di Kabupaten Halmahera Selatan

Poverty is one of the crucial issues faced by various countries including Indonesia. As a developing country, poverty is not a new problem. Almost all government periods in Indonesia have placed poverty as a development issue, especially South Halmahera Regency as one of the areas in North Maluku Province. Poverty in this area fluctuates every year. This study aims to analyze the effect of the Human Development Index (HDI), Unemployment, and Labor on poverty in South Halmahera Regency. This study is a quantitative study, the type of data used is secondary data in the form of time series data from 2010 - 2023 sourced from the Central Statistics Agency. The data collected was processed using the multiple linear regression analysis method with the help of Eviews version 12. The results of the study show that: (1) HDI partially has a negative and significant effect on poverty in South Halmahera Regency. (2) Unemployment partially has a negative and significant effect on poverty in South Halmahera Regency. (3) The workforce partially has a positive and significant influence on poverty in South Halmahera Regency. (4) The Human Development Index, Unemployment (HDI), and the workforce simultaneously influence poverty in South Halmahera Regency.

UnfoldSim.jl: Simulating continuous event-based time series data for EEG and beyond

UnfoldSim.jl: Simulating continuous event-based time series data for EEG and beyond

Developing a seasonal-adjusted machine-learning-based hybrid time‑series model to forecast heatwave warning

Heatwaves pose a significant threat to environmental sustainability and public health, particularly in vulnerable regions and rapidly growing cities. They cause water shortages, stress on plants, and an overall drying out of landscapes, reducing plant growth—the basis of energy production and the food chain. Accurate heatwave forecasting is crucial for early warning systems, public health interventions, and disaster preparedness strategies, reducing heat-related mortality risk through modeling and evaluation of warnings. However, anticipating heatwave warnings requires handling the daily time series data, which is a large-scale and high-frequency time series data. High-frequency time series data forecasting presents unique challenges due to its inherent complexity and characteristics. Therefore, the study proposes two algorithms to develop Machine-Learning (ML)-based hybrid models as well as seasonal adjusted ML-based hybrid models, which can handle large datasets and reveal complex seasonal patterns. The performance of these developed ML-based hybrid models and seasonal adjusted ML-based hybrid models were compared with other traditional time series, Autoregressive Integrated Moving Average (ARIMA), Exponential Smoothing State Space (ETS), and Trigonometric Box-Cox ARMA Trend Seasonal (TBATS) and ML models, Artificial Neural Network (ANN), Support Vector Regression (SVR), Prophet, Random Forest Regression (RFR), and Long Short-Term Memory (LSTM), to forecast heatwave warnings in Rajshahi, one of Bangladesh’s warmest districts, based on 42-year historical daily instances. Our findings indicate that the seasonal adjusted ML-based hybrid model, by integrating the Seasonal-Trend decomposition procedure based on LOESS (STL) approach with different time series and ML models, STL-ARIMA-LSTM, outperformed all other models with MAE (0.8974), MAPE (2.9232), RMSE (1.1794), MASE (0.3814) and ACF1 (0.0026). Hence, our suggested seasonal adjusted ML-based hybrid model, ensures a more accurate forecast and helps to determine the number and days of heatwaves, enabling people to plan ahead and take necessary safety measures before they occur.

A dual branch model for predicting microseismic magnitude time series named DTFNet

Microseismic monitoring is crucial in realizing intelligent early warning of coal mine rockbursts. Utilizing historical microseismic monitoring data to predict future microseismic events effectively enhances the accuracy of impact disaster prediction and early warning. Due to the complexity and nonlinearity of microseismic data, conventional time series prediction models struggle to forecast them accurately. Therefore, this paper proposes a microseismic time series prediction model, DTFNet, which integrates time series decomposition and deep learning. Initially, the original microseismic magnitude data is decomposed, reconstructed, and subjected to secondary decomposition using complementary ensemble empirical mode decomposition, permutation entropy, and variational mode decomposition. Subsequently, a dual branch time series prediction model is constructed, which effectively models the microseismic time series data and deeply extracts the features contained in the microseismic magnitude data. This paper uses microseismic monitoring catalogs from multiple working faces as the dataset to predict microseismic magnitudes. The model’s performance is evaluated using four metrics: mean squared error, mean absolute error, relative standard error, and root mean squared error. Experiments show that the proposed method effectively predicts the trend of microseismic magnitude changes and demonstrates good generalization and accuracy. Compared to several popular deep learning time series prediction models, DTFNet reduces the four evaluation metrics by an average of 23%, 18.1%, 11.1%, and 12%, respectively, showcasing a significant competitive advantage.

Temporal-TimesNet: a novel hybrid model for vessel traffic flow multi-step prediction

ABSTRACT This paper introduces the Temporal-TimesNet (T-TimesNet) model for predicting vessel traffic flow. Utilizing the Fast Fourier Transform (FFT), the model decomposes time series data to extract multi-periodicity features. A novel Temporal Information-Enhanced TCN (TIE-TCN) module captures complex time features, supplemented by an attention-based fusion mechanism that integrates continuity and causality from different modules. A fully connected network maps these features to produce predictions. Tested on AIS data from Fujiangsha, China, T-TimesNet outperforms models like KNN, SVR, ELM, LSTM, and CNN-LSTM, achieving significant reductions in MAE and MAPE by 6.35% and 6.05% respectively.

Application of the KA-Transformer model to early sepsis prediction: a hybrid network analysis based on time series data

BackgroundDue to the high mortality rate of sepsis, timely identification and intervention in the intensive care unit are essential to improve patient outcomes. Accurate prediction models are critical in addressing the challenge of early sepsis detection.MethodsIn this study, we designed a prediction model, KA-Transformer, based on time series data to achieve advance prediction of sepsis. The kernel attention mechanism is introduced in KA-Transformer, which significantly optimizes the problems of limited training samples, many parameters, and uneven sample distribution.ResultsBy inputting continuous time-series data, the model predicted the area under the wok characteristic curve for subjects with sepsis to be 0.962, 0.944, and 0.984 at the three prediction time points of 1, 6, and 12 h before the onset of sepsis, with accuracies of 92.3%, 93.9%, and 96.1%, respectively.ConclusionsThe experimental results show that KA-Transformer outperforms existing methods in terms of accuracy and generalization ability for sepsis prediction, and has the potential to achieve continuous prediction and improve the real-time and reliability of prediction.

Diagnostic analytics for the mixed Poisson INGARCH model with applications

In statistical diagnosis and sensitivity analysis, the local influence method plays a crucial role and is sometimes more advantageous than other methods. The mixed Poisson integer-valued generalized autoregressive conditional heteroscedastic (INGARCH) model is built on a flexible family of mixed Poisson distributions. It not only encompasses the negative binomial INGARCH model but also allows for the introduction of the Poisson-inverse Gaussian INGARCH model and the Poisson generalized hyperbolic secant INGARCH model. This paper applies the local influence analysis method to count time series data within the framework of the mixed Poisson INGARCH model. For parameter estimation, the Expectation-Maximization algorithm is utilized. In the context of local influence analysis, two global influence methods (generalized Cook distance and Q-distance) and four perturbations–case weights perturbation, data perturbation, additive perturbation, and scale perturbation–are considered to identify influential points. Finally, the feasibility and effectiveness of the proposed methods are demonstrated through simulations and analysis of a real data set.

Prediction of Track Irregularities in High-Speed Railways based on CEEMDAN-IPSO-LSTM Model

Uneven tracks affect the comfort and safety of train operations. Predicting the trend of track irregularities aids railway departments in issuing early warnings, thereby ensuring safe operations of trains. The inherent complex nonlinear and non-stationary nature of time series data derived from track irregularities challenges the prediction accuracy of existing methods. To tackle this challenge, this paper proposes a model based on CEEMDAN-IPSO-LSTM to predict the development trend of track irregularities. First, the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) algorithm is used to decompose the original data. This decomposition method offers enhanced completeness, faster decomposition, and the ability to capture richer information and structures within the original data. Subsequently, the decomposed multiple subsequences are input into a Long Short-Term Memory (LSTM) network model for prediction. This model is adept at handling long-term dependencies in time series data and can more accurately capture data relationships for achieving more precise predictions. In addition, the Improved Particle Swarm Optimization (IPSO) algorithm is used to optimize the hyperparameters of the LSTM model during training. Finally, the predicted subsequence results are reconstructed to derive the final prediction results. Comparative experiments with multiple models demonstrate the superior prediction accuracy of the CEEMDAN-IPSO-LSTM model proposed in this paper. Compared with the state-of-the-art Transformer, our model has achieved an average improvement of 15% in evaluation metrics such as RMSE, MAE, and R-squared.

Evaluation-free Time-series Forecasting Model Selection via Meta-learning

Time-series forecasting models are invariably used in a variety of domains for crucial decision-making. Traditionally these models are constructed by experts with considerable manual effort. Unfortunately, this approach has poor scalability while generating accurate forecasts for new datasets belonging to diverse applications. Without access to skilled domain-knowledge, one approach is to train all the models on the new time-series data and then select the best one. However, this approach is nonviable in practice. In this work, we develop techniques for fast automatic selection of the best forecasting model for a new unseen time-series dataset, without having to first train (or evaluate) all the models on the new time-series data to select the best one. In particular, we develop a forecasting meta-learning approach called AutoForecast that allows for the quick inference of the best time-series forecasting model for an unseen dataset. Our approach learns both forecasting models’ performances over time horizon of the same dataset and task similarity across different datasets. The experiments demonstrate the effectiveness of the approach over state-of-the-art (SOTA) single and ensemble methods and several SOTA meta-learners (adapted to our problem) in terms of selecting better forecasting models (i.e., 2 \(\times\) gain) for unseen tasks for univariate and multivariate testbeds. AutoForecast has also significant reduction in inference time compared to the naïve approach (doing inference using all possible models and then selecting the best one), with median of 42 \(\times\) across the two testbeds. We release our meta-learning database corpus (348 datasets), performances of the 322 forecasting models on the database corpus, meta-features, and source codes for the community to access them for forecasting model selection and to build on them with new datasets and models which can help advance automating time-series forecasting problem. In our released database corpus, we unveil new traces of Adobe computing cluster usage for production workloads.

Advanced Trajectory Analysis of NASA’s Juno Mission Using Unsupervised Machine Learning: Insights into Jupiter’s Orbital Dynamics

NASA’s Juno mission, involving a pioneering spacecraft the size of a basketball court, has been instrumental in observing Jupiter’s atmosphere and surface from orbit since it reached the intended orbit. Over its first decade of operation, Juno has provided unprecedented insights into the solar system’s origins through advanced remote sensing and technological innovations. This study focuses on change detection in terms of Juno’s trajectory, leveraging cutting-edge data computing techniques to analyze its orbital dynamics. Utilizing 3D position and velocity time series data from NASA, spanning 11 years and 5 months (August 2011 to January 2023), with 5.5 million samples at 1 min accuracy, we examine the spacecraft’s trajectory modifications. The instantaneous average acceleration, jerk, and snap are computed as approximations of the first, second, and third derivatives of velocity, respectively. The Hilbert transform is employed to visualize the spectral properties of Juno’s non-stationary 3D movement, enabling the detection of extreme events caused by varying forces. Two unsupervised machine learning algorithms, DBSCAN and OPTICS, are applied to cluster the sampling events in two 3D state spaces: (velocity, acceleration, jerk) and (acceleration, jerk, snap). Our results demonstrate that the OPTICS algorithm outperformed DBSCAN in terms of the outlier detection accuracy across all three operational phases (OP1, OP2, and OP3), achieving accuracies of 99.3%, 99.1%, and 98.9%, respectively. In contrast, DBSCAN yielded accuracies of 98.8%, 98.2%, and 97.4%. These findings highlight OPTICS as a more effective method for identifying outliers in elliptical orbit data, albeit with higher computational resource requirements and longer processing times. This study underscores the significance of advanced machine learning techniques in enhancing our understanding of complex orbital dynamics and their implications for planetary exploration.

The Application of Discrete Hilbert transform in Measuring Asset Return Risk in Investment Portfolios

AbstractPurpose: This study aims to explore the application of the discrete Hilbert transform (DHT) in measuring asset return risk within a portfolio, emphasizing its potential for enhancing risk evaluation in the field of financial mathematics.Methodology/approach: The research utilized a quantitative approach with data sourced from publicly available historical stock prices. Analysis was conducted using software MATLAB to implement the discrete Hilbert transform, which transforms time-series data into phase and amplitude components. The portfolio risk was calculated based on the transformed data, and the results were compared against traditional risk metrics such as variance and Value at Risk (VaR).Results/findings: The findings indicate that the discrete Hilbert transform provides additional insights into portfolio risk by capturing frequency-domain characteristics of asset returns. It complements traditional measures, offering a novel perspective on risk analysis. Specifically, the discrete Hilbert transform was effective in identifying subtle changes in risk patterns that were not apparent in time-domain analyses alone.Conclusion: This study investigates the application of the discrete Hilbert transform (DHT) in measuring the risk of asset returns within investment portfolios, specifically focusing on Indonesian stocks. TheLimitations: This study is limited by its focus on a small sample of stocks within a single financial market, which may restrict the generalizability of the findings. Additionally, the research does not account for macroeconomic factors that could influence asset returns.Contribution: This study contributes to the field of financial risk management by introducing the discrete Hilbert transform as a supplementary tool for risk analysis. It offers practical implications for portfolio managers, actuaries, and financial analysts seeking innovative methods to enhance risk assessment and decision-making processes.

Generative AI Models in Time-Varying Biomedical Data: Scoping Review

Background Trajectory modeling is a long-standing challenge in the application of computational methods to health care. In the age of big data, traditional statistical and machine learning methods do not achieve satisfactory results as they often fail to capture the complex underlying distributions of multimodal health data and long-term dependencies throughout medical histories. Recent advances in generative artificial intelligence (AI) have provided powerful tools to represent complex distributions and patterns with minimal underlying assumptions, with major impact in fields such as finance and environmental sciences, prompting researchers to apply these methods for disease modeling in health care. Objective While AI methods have proven powerful, their application in clinical practice remains limited due to their highly complex nature. The proliferation of AI algorithms also poses a significant challenge for nondevelopers to track and incorporate these advances into clinical research and application. In this paper, we introduce basic concepts in generative AI and discuss current algorithms and how they can be applied to health care for practitioners with little background in computer science. Methods We surveyed peer-reviewed papers on generative AI models with specific applications to time-series health data. Our search included single- and multimodal generative AI models that operated over structured and unstructured data, physiological waveforms, medical imaging, and multi-omics data. We introduce current generative AI methods, review their applications, and discuss their limitations and future directions in each data modality. Results We followed the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines and reviewed 155 articles on generative AI applications to time-series health care data across modalities. Furthermore, we offer a systematic framework for clinicians to easily identify suitable AI methods for their data and task at hand. Conclusions We reviewed and critiqued existing applications of generative AI to time-series health data with the aim of bridging the gap between computational methods and clinical application. We also identified the shortcomings of existing approaches and highlighted recent advances in generative AI that represent promising directions for health care modeling.

Time Series Remote Sensing Image Classification with a Data-Driven Active Deep Learning Approach

Recently, Time Series Remote Sensing Images (TSRSIs) have been proven to be a significant resource for land use/land cover (LULC) mapping. Deep learning methods perform well in managing and processing temporal dependencies and have shown remarkable advancements within this domain. Although deep learning methods have exhibited outstanding performance in classifying TSRSIs, they rely on enough labeled time series samples for effective training. Labeling data with a wide geographical range and a long time span is highly time-consuming and labor-intensive. Active learning (AL) is a promising method of selecting the most informative data for labeling to save human labeling efforts. It has been widely applied in the remote sensing community, except for the classification of TSRSIs. The main challenge of AL in TSRSI classification is dealing with the internal temporal dependencies within TSRSIs and evaluating the informativeness of unlabeled time series data. In this paper, we propose a data-driven active deep learning framework for TSRSI classification to address the problem of limited labeled time series samples. First, a temporal classifier for TSRSI classification tasks is designed. Next, we propose an effective active learning method to select informative time series samples for labeling, which considers representativeness and uncertainty. For representativeness, we use the K-shape method to cluster time series data. For uncertainty, we construct an auxiliary deep network to evaluate the uncertainty of unlabeled data. The features with rich temporal information in the classifier’s middle-hidden layers will be fed into the auxiliary deep network. Then, we define a new loss function with the aim of improving the deep model’s performance. Finally, the proposed method in this paper was verified on two TSRSI datasets. The results demonstrate a significant advantage of our method over other approaches to TSRSI. On the MUDS dataset, when the initial number of samples was 100 after our method selected and labeled 2000 samples, an accuracy improvement of 4.92% was achieved. On the DynamicEarthNet dataset, when the initial number of samples was 1000 after our method selected and labeled 2000 samples, an accuracy improvement of 7.81% was attained. On the PASTIS dataset, when the initial number of samples was 1000 after our method selected and labeled 2000 samples, an accuracy improvement of 4.89% was achieved. Our code is available in Data Availability Statement.

A dual-channel transferable model for cross-domain remaining useful life prediction of rolling bearings under uncertainty

Abstract Accurate prediction of the remaining useful life (RUL) of rolling bearings is a challenging task due to the complexity of time series data, inherent uncertainty in predictions, and domain shift between source and target datasets. To address these challenges, a parallel deep learning framework is proposed that integrates transfer learning and uncertainty quantification for RUL prediction. The framework combines AlexNet with the convolutional block attention module and Transformer with a gated convolutional unit to effectively extract degradation features from time series data. Additionally, Bayesian optimization is employed for hyperparameter tuning, reducing the need for manual adjustments, while a domain adaptation module addresses distribution differences between the source and target domains. Variational inference methods extend the model to a Bayesian deep neural network, providing confidence intervals for uncertainty quantification in RUL predictions. Experimental results on two rolling bearing datasets demonstrate that the proposed CANN-GT-BDA model outperforms state-of-the-art models in terms of predictive accuracy, offering a reliable and uncertainty-aware solution for industrial predictive maintenance.