A0650
Title: One class classification using Bayesian optimization
Authors: In Young Baek - Inha University (Korea, South) [presenting]
Seongil Jo - Inha University (Korea, South)
Jae Oh Kim - Inha University (Korea, South)
Abstract: One Class Classification(OCC) is a technique for detecting abnormal data by creating a decision boundary that defines normal data when there is an imbalance between normal and abnormal data. One Class Support Vector Machine (OC-SVM) and Deep Support Vector Data Description (Deep SVDD) are one of the methodologies of OCC. OCSVM is to find the hyperplane that separates the majority of normal data from the origin in the feature space. Deep SVDD is to find the smallest hypersphere involving the most normal data. By using neural networks, Deep SVDD maps data from the original space to a feature space. OCSVM and Deep SVDD are sensitive to hyperparameters. Methodologies of hyperparameters optimization are Grid Search, Random Search and Bayesian Optimization. The end is to compare the performance between the three methodologies of hyperparameter optimization and show that Bayesian optimization outperforms grid search and random search.
