Adaptive Memory Broad Learning System for Unsupervised Time Series Anomaly Detection.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Time series anomaly detection is crucial for identifying unusual patterns.
• Existing methods face challenges in understanding time-independent and abnormal data characteristics.

Purpose of the Study:

• To propose a novel algorithm, adaptive memory broad learning system (AdaMemBLS), for effective time series anomaly detection.
• To enhance the model's ability to learn time series data characteristics and improve anomaly detection.

Main Methods:

• Leveraging the broad learning algorithm's rapid inference and a memory bank for data differentiation.
• Implementing an incremental algorithm with multiple data augmentation techniques for ensemble learners.
• Utilizing a diverse ensemble approach and a discriminative anomaly score.

Main Results:

• The proposed AdaMemBLS method demonstrates superior inference speed.
• Achieved more accurate anomaly detection compared to existing competitive methods.
• Extensive experiments on real-world datasets validate the method's efficacy.

Conclusions:

• AdaMemBLS provides an effective and efficient solution for time series anomaly detection.
• The combination of broad learning, memory banks, and ensemble techniques enhances performance.
• The study offers a detailed investigation into the method's effectiveness and underlying principles.