Efficient SVDD sampling with approximation guarantees for the decision boundary

Abstract Support Vector Data Description (SVDD) is a popular one-class classifier for anomaly and novelty detection. But despite its effectiveness, SVDD does not scale well with data size. To avoid prohibitive training times, sampling methods select small subsets of the on which trains decision boundary hopefully equivalent to one obtained full set. According literature, good sample should therefore contain so-called observations that would as support vectors However, non-boundary also are essential fragment contiguous inlier regions poor classification accuracy. Other aspects, such selecting sufficiently representative sample, important well. existing largely overlook them, resulting in In this article, we study how considering these points. Our approach frame an optimization problem, where constraints guarantee indeed approximates original boundary. We then propose RAPID, efficient algorithm solve problem. RAPID require any tuning parameters, easy implement scales large sets. evaluate our real-world synthetic data. evaluation most comprehensive so far. results show outperforms competitors accuracy, size, runtime.