A hybrid rule extraction method for One-Class Support Vector Machines

One-Class Support Vector Machines (OCSVM) is widely applied for those classification problems where one of the classes of data is completely not present or not properly sampled. However, the knowledge presented by OCSVM is not interpretable by the human analyst. To ameliorate this problem there is a need for providing explanation aids to OCSVM classification decisions. Recently, several rule extraction methods were evolved to provide explanation ability to Support Vector Machines (SVM) and Artificial Neural Networks (ANN). Motivated from these methods this paper proposes a new hybrid rule extraction method for OCSVM, which is not widely studied yet. Proposed hybrid is composed with Support Vector Data Description (SVDD) and RIPPER rule learning. The viability of the proposed hybrid is tested over three benchmark datasets. Obtained results have shown that SVDD+RIPPER hybrid outperformed SVDD and RIPPER alone in terms of classification performance and explanation ability.