Hyperparameter Selection for Gaussian Process One-Class Classification.

Area of Science:

• Machine Learning
• Statistical Modeling
• Pattern Recognition

Background:

• Gaussian processes (GPs) offer robust statistical descriptions for predictions, enabling confidence intervals and hyperparameter tuning.
• Traditional GP hyperparameter selection methods are insufficient for the unique demands of one-class classification (OCC).

Purpose of the Study:

• To propose an effective method for selecting hyperparameters specifically for Gaussian process one-class classification (GP OCC).
• To address the limitations of standard GP hyperparameter tuning in the context of OCC.

Main Methods:

• The proposed method involves calculating the prediction difference between edge and interior positive training samples.
• This difference is utilized to optimize hyperparameters for GP OCC models.

Main Results:

• Experiments on 2-D artificial datasets demonstrated the method's effectiveness.
• Validation on University of California benchmark datasets confirmed the proposed approach's efficacy in GP OCC.

Conclusions:

• The novel hyperparameter selection strategy is effective for Gaussian process one-class classification.
• This method enhances the performance of GPs in OCC tasks by addressing hyperparameter tuning challenges.