Outlier detection of clustered functional data with image and signal processing applications by archetype analysis

Area of Science:

• Data Science
• Machine Learning
• Statistics

Background:

• Anomaly detection is crucial for identifying unusual patterns in data.
• Existing methods often struggle with detecting outliers in clustered functional datasets.
• Multivariate analysis techniques need adaptation for functional data contexts.

Purpose of the Study:

• To develop an innovative methodology for anomaly detection of curves.
• To extend the AA + kNN technique for application to functional data.
• To identify outliers within clustered functional data sets.

Main Methods:

• Extension of the AA + kNN technique to functional data.
• Comparative analysis against twelve state-of-the-art anomaly detection methods.
• Validation through simulated data with single and multiple functional clusters.

Main Results:

• The proposed method demonstrates superior performance in anomaly detection for functional data.
• Effectiveness validated across simulated clustered datasets.
• Successful application in computer vision and signal processing tasks.

Conclusions:

• The novel methodology provides an effective solution for anomaly detection in clustered functional data.
• The extended AA + kNN approach offers improved outlier identification capabilities.
• The research facilitates further advancements in functional data analysis and anomaly detection.