Boosted RVM Algorithm for Imbalanced and Noisy Data

Abstract

Relevance Vector Machine (RVM) is a machine learning algorithm based on sparse Bayesian theory, which performs good classification performance for small-scale data sets. However, it shows sub-optimal performance on noisy and imbalanced data sets. Therefore, a boosted RVM algorithm based on hybrid sampling and noise-detection (HSNDB-RVM) is proposed. In this approach, the imbalanced degree of samples is reduced by adopting a hybrid sampling method, and the impact of noise on the boosted RVM algorithm is weaken by decreasing the weight of noisy samples detected on top of RVM. The algorithm was applied to real data sets, and experimental results show that the proposed method offers good performance and can improve the ability of the RVM to process imbalanced and noisy data sets.