Rule Extraction from Support Vector Machines and Its Applications

Support Vector Machines are the state-of-the-art tools in data mining. However, their strength are also their main weakness, as the generated nonlinear models are typically regarded as incomprehensible black-box models. Therefore, opening the black-boxor making SVMs explainable became more important and necessary in areas such as medical diagnosis and credit evaluation. Rule extraction from SVMs, which is in order to make SVMs more explainable has developed during recent years. However, existing rule extracted algorithms have limitations in real applications especially when the problems are large scale with high dimensions. In this paper, we combined two feature selection techniques with rule extraction from SVMs in order to deal with this case. And we also proposed a new criteria to evaluate the extracted rules in order to rich the evaluation standards. Numerical experiments show the efficiency of our method.