FuseAD: Unsupervised Anomaly Detection in Streaming Sensors Data by Fusing Statistical and Deep Learning Models.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• The proliferation of smart devices generates vast amounts of streaming data from diverse sensors.
• Effective unsupervised anomaly detection is crucial for minimizing machine downtime through continuous monitoring.
• Existing anomaly detection methods, including statistical and deep learning techniques, have limitations depending on data type and use-case.

Purpose of the Study:

• To introduce FuseAD, a novel unsupervised anomaly detection technique that integrates statistical and deep learning methods.
• To evaluate the performance of FuseAD against state-of-the-art anomaly detection approaches.
• To demonstrate the benefits of a fusion-based strategy in anomaly detection.

Main Methods:

• Developed FuseAD, a hybrid technique fusing statistical (ARIMA) and deep learning (CNN) models in a residual manner.
• Tested FuseAD on a publicly available dataset (Yahoo Webscope benchmark).
• Conducted an ablation study to assess the contribution of individual components within FuseAD.

Main Results:

• FuseAD demonstrated improved performance, indicated by an increased Area Under the Curve (AUC), compared to existing state-of-the-art methods.
• The fusion approach effectively combines the strengths of both statistical and deep learning models.
• The ablation study confirmed the significant contributions of both the ARIMA and CNN components.

Conclusions:

• The proposed FuseAD technique offers a robust solution for unsupervised anomaly detection in streaming data.
• Hybrid approaches combining statistical and deep learning methods can outperform individual techniques.
• FuseAD provides a promising direction for enhancing the reliability and efficiency of machine monitoring systems.