Creating and measuring diversity in multiple classifier systems using support vector data description

In this paper, a new method is introduced to construct Multiple Classifier Systems (MCSs). It is based on controlling diversity among base classifiers according to a new method in measuring diversity in kernel space. The method admits a tradeoff between individual classifier and multiple classifier accuracy and diversity as each base classifier requires knowledge of the choices made by the other MCS members. This knowledge is included in the method using data descriptors as a tool for creating diversity between base classifiers in kernel space. Data description properties are also used for measuring diversity. A new combining method presented in this paper completes this work.Performance of the proposed method is evaluated on a number of known benchmark datasets. Analyzing the results shows that the proposed method improves system's overall performance and accuracy in many cases. It also measures diversity more precisely.