Multiclass classification based on extended support vector data description.

Area of Science:

• Machine Learning
• Pattern Recognition
• Data Mining

Background:

• Support Vector Data Description (SVDD) is a one-class classification method.
• Existing SVDD methods may struggle with complex datasets and negative sample integration.
• Effective classification is crucial for applications like industrial monitoring.

Purpose of the Study:

• To propose two new variations of SVDD incorporating negative samples (NSVDD).
• To extend NSVDD for multiclass classification using Linear Discriminant Analysis (LDA) and Nearest Neighbor (NN) rule.
• To evaluate the proposed methods on benchmark datasets and a real-world roller bearing monitoring task.

Main Methods:

• Development of two-norm NSVDD and nu-NSVDD algorithms.
• Integration of LDA and NN rule for multiclass classification in kernel feature space.
• Extensive simulations using eight benchmark datasets and a real-world vibration signal dataset.

Main Results:

• Proposed NSVDD methods demonstrated lower classification error rates and standard deviations.
• The two-norm NSVDD with LDA-NN achieved 100.0% accuracy for binary fault detection in roller bearings.
• The method achieved 99.9% accuracy for multiclass classification of roller bearings under six conditions.

Conclusions:

• The proposed NSVDD variations offer improved classification performance over traditional methods.
• The integration with LDA-NN effectively extends NSVDD to multiclass problems.
• The methods show significant promise for real-world applications such as condition monitoring.