Concept Drift Research Articles (Page 1)

Abstract The rapid proliferation of images on online platforms has made emotion analysis a task of paramount significance. However, these images are often privacy-sensitive, making Federated Learning (FL) a compelling paradigm over traditional centralized methods. A critical yet largely unaddressed challenge in applying FL to this domain is the severe concept drift stemming from the subjective and culturally diverse nature of emotional expression, which causes conventional FL algorithms to fail. In this paper, we propose CAFL (Conditional Attention Federated Learning) to fill this gap. CAFL empowers clients to learn collaboratively yet personally. It intelligently routes information through an adaptive gate that separates features into a personalized stream and a global stream. These streams are then processed by dedicated local and global prediction heads. Crucially, collaboration is guided by a conditional attention mechanism, where the server computes a personalized reference model for each client based on an attention-weighted aggregation of peer models, promoting knowledge sharing among kindred clients. Extensive experiments on various lightweight foundation models show that CAFL consistently outperforms existing FL methods, demonstrating its robustness and superior performance as a solution for distributed, privacy-sensitive image emotion analysis.

Machine Learning models are often trained on historical data with the assumption that the future data will follow similar pattern. However, in real world applications data shifts over time due to factors such as change in market conditions and user behaviour, resulting in drift. Drift leads to the degradation of model performance, making its management critical for reliable predictions in dynamic environments. This study offers a thorough examination of types of drift and existing detection techniques, including statistical and machine learning based approach. Furthermore, the research will examine the significance drift in cybersecurity, where adaptive strategies introduce both data and concept drift. The aim of this paper is to bridge the gap between machine learning theories with practical security applications.

The rapid growth of AI and data science has made streaming analysis crucial in finance, healthcare, IoT, and transportation, where massive multimodal data are continuously generated. In such settings, concept drift arises within a single modality and from cross-modal consistency shifts, which can degrade model performance. This paper reviews recent progress on multimodal concept drift and defines two categories: intra-modal distributional and inter-modal consistency change. We propose three main contributions: cross-modal canonical correlation analysis, drift detection with deep embedding learning, and incremental adaptation via knowledge distillation. Experiments show that our multimodal approach detects drift more accurately and with less delay than single-modality methods, while avoiding catastrophic forgetting. It also improves adaptation efficiency by about 70% compared with full retraining. These results demonstrate the value of integrating drift detection and adaptive learning to sustain long-term performance in multimodal environments.

Financial fraud represents a critical global challenge with substantial economic and social consequences. This comprehensive review synthesizes the current knowledge on machine learning approaches for financial fraud detection, examining their effectiveness across diverse fraud scenarios. We analyze various fraud types, including credit card fraud, financial statement fraud, insurance fraud, and money laundering, along with their specific detection challenges. The review outlines supervised, unsupervised, and hybrid learning approaches, discussing their applications and performance in different fraud detection contexts. We examine commonly used datasets in fraud detection research and evaluate performance metrics for assessing these systems. The review is further grounded by two case studies applying supervised models to real-world banking data, illustrating the practical challenges of implementing fraud detection systems in operational environments. Through our analysis of the recent literature, we identify persistent challenges, including data imbalance, concept drift, and privacy concerns, while highlighting the emerging trends in deep learning and ensemble methods. This review provides valuable insights for researchers, financial institutions, and practitioners working to develop more effective, adaptive, and interpretable fraud detection systems capable of operating within real-world financial environments.

BackgroundLarge language models (LLMs) are increasingly being applied in healthcare; however, their performance in specialized fields, such as oncology, is subject to temporal factors, including knowledge decay and concept drift. The impact of these temporal dynamics on LLM question-answering accuracy in oncology remains inadequately evaluated. This study aims to systematically assess the temporal evolution of LLM accuracy in responding to oncology-related questions using real-world data.MethodWe systematically collected relevant literature through 2025 by searching LLM-related keywords in PubMed, Google Scholar, and Web of Science databases. The inclusion criteria were as follows: (1) cancer-related research; (2) clear and complete question descriptions; and (3) complete answers. The final sample (n = 23) contained 614 research questions, comprising subjective questions (n = 223) and multiple-choice questions (n = 391). Following randomization of responses generated by three LLMs (ChatGPT-3.5, ChatGPT-4, and Gemini), we evaluated their accuracy across different cancer categories using both original scoring criteria and Likert scale scoring methods. Data analysis was performed using R statistical software, employing random or fixed effects models to calculate pooled mean differences (MD) and relative risks (RR) with their 95% confidence intervals (CI).ResultsThe findings demonstrated that in both subjective and objective oncology assessments, ChatGPT-3.5 (subjective questions MD = −3.30; objective questions RR = 0.92) and ChatGPT-4 (subjective questions MD = −7.17; objective questions RR = 0.93) showed declining performance trends over time, while Gemini exhibited significant improvements over time (subjective questions MD = 11.48; objective questions RR = 1.15). Notably, ChatGPT-3.5‘s performance on subjective questions revealed a significant turning point between March 14, 2023, and April 26, 2023, shifting from initially superior performance on newer questions to inferior performance compared with original questions, with the performance gap progressively widening.ConclusionsOur meta-analysis reveals temporal performance degradation in ChatGPT-3.5 and ChatGPT-4, which contrasts with the consistent improvement observed in Gemini. These findings provide essential guidance for the evidence-based deployment of LLMs in oncology.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12967-025-07227-2.

Phishing detection models degrade quickly due to drift, adversarial evasion, and fairness issues. Existing MLOps platforms mainly automate deployment and monitoring. Prior works have examined SHAP-based monitoring, retraining, or fairness audits separately, but lack an integrated theory of resilience for adversarial environments. We introduce the Hybrid MLOps Framework (HAMF), a system designed to embed resilience and ethical governance into the lifecycle of phishing detection models. HAMF is ‘hybrid’ because it unifies proactive and reactive adaptation, combining automation with stakeholder oversight, and embedding resilience with ethical governance. HAMF treats resilience as an integrated lifecycle property, designed to simultaneously preserve model accuracy, fairness, and stakeholder trust amidst concept drift. Methodologically, HAMF implements this through a hybrid control cycle. This cycle fuses four key capabilities: SHAP-guided feature replacement, event-driven retraining, fairness-triggered audits, and structured human feedback. Unlike conventional pipelines where these functions are isolated, HAMF ensures their interdependence as first-class triggers. Empirical evaluations on large-scale phishing streams demonstrate HAMF’s superior performance. The framework detects drift within 18 seconds, restores F1 scores above 0.99 post-attack, reduces subgroup disparities by over 60%, and scales to over 2,300 requests per second with sub-50ms latency. These results validate HAMF’s design, demonstrating that embedding resilience and ethical alignment into the MLOps lifecycle is both effective and scalable.

The digital economy’s increasing integration with financial systems has amplified systemic vulnerabilities, yet most early warning frameworks remain static and slow to adapt to evolving market, macroeconomic, and sentiment dynamics. This study develops an AI-driven early warning system that fuses macroeconomic indicators, market data from major digital firms, and online sentiment metrics to detect and anticipate financial stress within the U.S. digital economy. Leveraging a decade of weekly data (2015–2025) across FRED, Yahoo Finance, Reddit, and Google News sources, we construct a multimodal dataset capturing both structural and behavioral signals. A hybrid modeling pipeline integrates interpretable Logistic Regression with adaptive online learning using the River framework and ADWIN drift detection, enabling real-time updates as financial conditions evolve. Experiments benchmark Logistic Regression, Random Forest, and XGBoost against the adaptive model under temporal cross-validation. Results show that the interpretable logistic regression baseline achieves the most stable performance across folds, while the adaptive River model sustains predictive capability under concept drift, providing continuous calibration without retraining. Lead-time analysis demonstrates the system’s ability to issue warnings 2–6 weeks before major stress events, validated through both macroeconomic and sentiment-driven market shifts. Explainability analysis using SHAP reveals that negative sentiment dynamics, rising volatility, and tightening monetary conditions are the strongest precursors of stress. These findings highlight how transparent, adaptive AI systems can bridge the gap between accuracy, interpretability, and operational feasibility, offering a practical foundation for real-time systemic risk monitoring in digital financial ecosystems.

Vibration-based predictive maintenance is an essential element of reliability engineering for modern automotive powertrains including internal combustion engines, hybrids, and battery-electric platforms. This review synthesizes advances in sensing, signal processing, and artificial intelligence that convert raw vibration into diagnostics and prognostics. It characterizes vibration signatures unique to engines, transmissions, e-axles, and power electronics, emphasizing order analysis, demodulation, and time–frequency methods that extract weak, non-stationary fault content under real driving conditions. It surveys data acquisition, piezoelectric and MEMS accelerometry, edge-resident preprocessing, and fleet telemetry, and details feature engineering pipelines with classical machine learning and deep architectures for fault detection and remaining useful life prediction. In contrast to earlier reviews focused mainly on stationary industrial systems, this review unifies vibration analysis across combustion, hybrid, and electric vehicles and connects physics-based preprocessing to scalable edge and cloud implementations. Case studies show that this integrated perspective enables practical deployment, where physics-guided preprocessing with lightweight models supports robust on-vehicle inference, while cloud-based learning provides cross-fleet generalization and model governance. Open challenges include disentangling overlapping sources in compact e-axles, coping with domain and concept drift from duty cycles, software updates, and aging, addressing data scarcity through augmentation, transfer, and few-shot learning, integrating digital twins and multimodal fusion of vibration, current, thermal, and acoustic data, and deploying scalable cloud and edge AI with transparent governance. By emphasizing inverter-aware analysis, drift management, and benchmark standardization, this review uniquely positions vibration-based predictive maintenance as a foundation for next-generation vehicle reliability.

Research on adaptive online anomaly detection for satellite sensor data under concept drift and missing values

Self-supervise ensemble for extreme imbalance data streams with concept drift

IncA-DES: An incremental and adaptive dynamic ensemble selection approach using online K-d tree neighborhood search for data streams with concept drift

TEDA-driven adaptive stream clustering for concept drift detection

TD-IVDM: A multi-scale concept drift detection method for time series forecasting tasks

STE-HOLNet: A new method for wind power prediction by integrating spatio-temporal features, dynamic concept drift detection and adaptive correction

The rapid evolution of cyber threats poses significant challenges to the adaptability and performance of anomaly detection systems. This study presents an innovative hybrid deep learning framework that integrates Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), and Transformer models with a novel self-learning mechanism to enhance network traffic anomaly detection. Our key contributions include: (1) a synergistic two-stage model fusion architecture that captures both spatial and temporal traffic patterns; (2) an adaptive learning mechanism with multi-metric drift detection that autonomously responds to evolving threats; and (3) a knowledge preservation strategy that maintains detection capabilities while adapting to new attack patterns. The proposed CNN-LSTM model achieves F1-scores of 0.9778 and 0.9695 on the UNSW-NB15 and CICIDS2017 datasets respectively for binary classification of normal vs. anomalous traffic. The LSTM-Transformer model further classifies specific anomaly types with accuracies of 0.9632 and 0.9528 on these datasets, representing significant improvements over recent methods. Experiments demonstrate the framework’s robustness, maintaining an average accuracy of 0.955 ( 0.005) over a 15-day simulated period with multiple induced concept drifts. The self-learning mechanism, with multi-metric drift detection and an efficient model update strategy, enables the system to detect drifts and recover performance within 23.4 ± 0.20 hours post-drift, while achieving a 92.8% detection rate for zero-day attacks. The proposed framework offers a promising direction for developing efficient and autonomous cybersecurity systems capable of handling dynamic and evolving threat landscapes.

Real-time IoT applications that use sensors, and those sensors generate continuous and unreliable data streams. This IoT stream suffers from Concept drift, which is defined as changes in data distribution over time. Traditional machine learning models fail to adapt to evolving environments, which degrades the model's performance. This paper presents a novel Weighted Ensemble Extreme Learning Machine (ELM) framework that integrates drift prediction, detection, and adaptation into a single architecture. The proposed approach employs an Adaptive Windowing (ADWIN) technique combined with non-parametric statistical tests, where the Kolmogorov-Smirnov (KS) test demonstrates superior performance in drift detection. To handle adaptation, an Online Sequential ELM ensemble is dynamically updated through exponential moving average-based weighting and pruning, ensuring robust generalization and reduced computational overhead. The framework is validated on benchmark datasets, including SEA, Stagger, and a farm weather dataset with artificially induced sudden, gradual, and recurring drifts using Gaussian noise. Experimental results reveal that the proposed method consistently outperforms single classifiers, achieving higher accuracy, precision, recall, and F1-scores, with an average accuracy improvement of up to 97.35% when drift handling is incorporated. These results demonstrate the effectiveness of combining ensemble learning with adaptive drift management. The contributions of this work lie in developing a comprehensive and lightweight framework that unifies prediction, detection, and adaptation of concept drift, thereby enhancing the reliability of IoT-driven decision-making systems. This research provides a scalable foundation for real-time applications such as weather forecasting and can be extended to other domains where streaming data is subject to dynamic changes.

Introduction: This study addresses the challenge of concept drift in multivariate time series (MTS) forecasting, where data distributions evolve, degrading model performance. The objective is to propose an adaptive hybrid model that combines Long Short-Term Memory (LSTM) networks with the ADaptive WINdowing (ADWIN) algorithm for effective drift detection and adaptive forecasting in real-time environments. Methods: The proposed ADWIN-LSTM framework integrates Bi-LSTM layers for temporal sequence modelling and ADWIN for monitoring prediction residuals to detect concept drift. A dynamic sliding window mechanism adjusts the training data scope based on drift type. Data preprocessing includes normalization, STL-based detrending, and PCA for dimensionality reduction. Hyperparameters are optimized using grid search. Results: Experiments on real-world and synthetic MTS datasets demonstrate that the proposed model outperforms baseline models (LSTM, GRU, CNN-LSTM, and ADWIN-RF) across RMSE, MAE, MAPE, and R² metrics. The model detects both abrupt and gradual drifts with minimal false positives and low detection delay (≤ 5 steps). Post-drift adaptation significantly improves forecasting accuracy. Discussion: The adaptive retraining strategy triggered by drift detection ensures computational efficiency and robustness in volatile environments. The dynamic integration of forecasting and drift detection enhances model adaptability to evolving data distributions. Conclusion: The ADWIN-LSTM framework effectively combines predictive learning and realtime drift adaptation, making it suitable for dynamic, high-stakes environments such as energy, traffic, and environmental systems. Future work includes online optimization and deployment on streaming platforms.

A global commitment has been observed toward the incorporation of experiential methodologies into education. Experiential learning strategies have been increasingly integrated into classrooms across various educational levels, and their use has been recognized as a core competency for educators (Council of Europe, 2018). The primary aim of this study is to analyze academic research concerning the application of experiential learning in music education. A sample was obtained from the Web of Science Core Collection, encompassing publications from January 1, 2014, to July 20, 2025. Multiple bibliometric tools, including GraphPad Prism v8.0.2, CiteSpace (6.2.4R), and VOSviewer (1.6.18), were employed to examine publication trends, relevant journals and authors, geographic distribution, keywords, and emerging research themes. The study analyzed 817 relevant publications spanning 76 countries and regions, 1,140 institutions, and 2,865 authors. The analysis demonstrated that (1) publication volume has shown a consistent upward trajectory, accelerating post-2019 with a peak anticipated in 2024; (2) the United States has led in both publication count (310, 37.94%) and citation frequency (5,111), followed by China; (3) the Journal of Chemical Education has been identified as the most prolific journal; and (4) research focal points have shifted from foundational topics such as categorization and integrated learning development to contemporary themes including active learning and concept drift. Looking ahead, two prominent areas of focus are anticipated to involve the integration of artificial intelligence and neural networks in music education, and the amalgamation of experiential learning with innovative pedagogical strategies.

Objectives: This paper introduces an adaptive learning framework for handling concept drift in data by dynamically adjusting model updates based on the severity of detected drift. Methods: The proposed method combines multiple statistical measures to quantify distributional changes between recent and historical data windows. The resulting severity score drives a three-tier adaptation policy: minor drift is ignored, moderate drift triggers incremental model updates, and severe drift initiates full model retraining. Results: This approach balances stability and adaptability, reducing unnecessary computation while preserving model accuracy. The framework is applicable to both single-model and ensemble-based systems, offering a flexible and efficient solution for real-time drift management. Also, different transformation methods were reviewed, and quantile transformation was tested. By applying a quantile transformation, the Kolmogorov–Smirnov (KS) statistic decreased from 0.0559 to 0.0072, demonstrating effective drift adaptation.

Digital businesses today face unprecedented hurdles in ensuring data quality in distributed systems, where conventional validation techniques prove unable to meet the speed and sophistication of modern information streams. The Real-Time Data Integrity Nexus prescribes a groundbreaking human-AI collaborative paradigm that aims to revolutionize autonomous data quality assurance by strategic fusion of machine learning innovations, event-driven design paradigms, and cloud-native orchestration frameworks. The framework creates synergies among artificial intelligence elements and human knowledge to produce adaptive surveillance systems that can identify anomalies in milliseconds while sustaining context awareness necessary for mission-critical systems. Stream processing architectures provide a continuous nice guarantee for petabyte-scale recordsets with discretized stream processing and fault-tolerant computing paradigms, ensuring reliable operation under first-rate load situations. Interactive gadget mastering procedures allow real-time model updates by means of human-in-the-loop comments, attaining higher performance than solely automated options without sacrificing interpretability and accountability. Advanced concept drift detection methods and data privacy protection technologies are supported for handling changing data distributions and compliance with regulatory needs. Horizontal scaling across thousands of computation nodes is supported by container orchestration technologies, while reinforcement learning components seek to optimize intervention tactics with ongoing adaptation. The architecture shows transformative value for autonomous quality assurance by synergizing human strategic control with machine computational power, creating new paradigms for data integrity management in real-time distributed environments that require both precision and responsiveness.