Robust Support Vector Data Description with Truncated Loss Function for Outliers Depression.

Area of Science:

• Machine Learning
• Data Mining
• Anomaly Detection

Background:

• Support Vector Data Description (SVDD) is a key method for anomaly detection.
• SVDD performance degrades with noisy or mislabeled training data.
• Robust anomaly detection methods are crucial for reliable data analysis.

Purpose of the Study:

• To develop a robust Support Vector Data Description (SVDD) model resistant to outliers and mislabeled data.
• To introduce a universal truncated loss function framework for SVDD.
• To enhance the generalization capabilities of SVDD in noisy environments.

Main Methods:

• Implemented a universal truncated loss function framework within the SVDD model.
• Utilized the fast alternating direction method of multipliers (ADMM) algorithm for solving truncated loss functions.
• Developed and analyzed truncated generalized ramp, binary cross entropy, and linear exponential loss functions for SVDD.
• Theoretically analyzed the convergence of the fast ADMM algorithm.

Main Results:

• The proposed truncated loss functions significantly improve SVDD robustness against data noise and outliers.
• Experimental results on synthetic and real-world datasets confirm the superior performance of the new SVDD models.
• The developed models exhibit enhanced generalization capabilities compared to standard SVDD.

Conclusions:

• The truncated loss function framework effectively enhances SVDD robustness for anomaly detection.
• The fast ADMM algorithm provides an efficient solution for these robust SVDD models.
• These novel SVDD models offer improved reliability in real-world applications with imperfect data.