Interpretable Sensor Change Detection via Conditional Cauchy-Schwarz Divergence

Area of Science:

• Data Science
• Machine Learning
• Signal Processing

Background:

• Multivariate sensor networks are crucial for monitoring complex systems.
• Existing change-point detection methods often lack interpretability, limiting their use in critical applications.

Purpose of the Study:

• To develop an interpretable framework for detecting distributional changes in multivariate sensor networks.
• To provide sensor-level explanations for detected changes.

Main Methods:

• The proposed method extends Cauchy-Schwarz (CS) divergence to conditional distributions.
• It detects distributional shifts by analyzing changes in sensor-wise conditional relationships.

Main Results:

• The framework achieves competitive or superior detection performance on synthetic and real-world data.
• It provides fine-grained interpretability for practical sensor monitoring.

Conclusions:

• The CS divergence-based method offers a reliable and interpretable solution for change detection in sensor networks.
• This approach enhances the practical utility of monitoring systems in safety-critical domains.