.png){kind=icon}
.png){kind=icon}
Return
Enhanced Approaches for Anomaly Detection in Streaming Data: Coupling Gaussian Distributions With Space Trees and Adaptive AutoEncoders
.png){kind=icon}
PRE
AI
E
Z
H
DOI：10.1109/tbdata.2026.3679567.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Processing, analyzing and continuously monitoring big data streams in real time is crucial for the security and efficiency of organizations and individuals. While machine learning and deep learning have greatly improved anomaly detection, they also have drawbacks, such as reliance on offline data and poor performance, especially with scarce or fluctuating data. To address this problem, we propose two enhanced methods for anomaly detection in streaming data: Gaussian Space Trees (GSTrees) and the Gaussian Weighted Adwin AutoEncoder (GWAAE). The effectiveness of these methods was evaluated on real datasets (ECG5000, Credit Card Fraud Detection, SMTP) processed as streaming data. For the ECG5000 dataset, GSTrees performed excellently on all metrics, consistently achieving over 94%, while GWAAE excelled with an ROC-AUC of over 89% and over 84% on all other metrics. For the SMTP dataset, the ROC-AUC of both proposed methods is above 80%. GSTrees and GWAAE successfully identified anomalies with a recall of over 82% and a ROC-AUC of over 94% and maintained this success with remarkably low false negative and false positive rates in the Credit Card Fraud Detection dataset. These results emphasize the robustness and applicability of GSTrees and GWAAE for real-time anomaly detection in dynamic environments.
Keywords:
Streaming data
online machine learning
online anomaly detection
autoencoder
unsupervised learning
Journal
I
IF:
5.7
Papers:
834
Citations:
3.0K
Organization
Cited Papers