Multi-Objective Optimization of SVM Parameters for Meta Classifier Design

ABSTRACT Support Vector Machines (SVM) have proved to be one of the most popular techniques for pattern classification that has been widely employed in many real-world application areas. The classification performance of the SVM largely depends on the kernel parameters setting and carefully tuning of the kernel parameters of SVM predominantly helps in the improvement in the classification performance. In literature, it is evident that tuning the kernel parameters has been formulated as optimization problems and evolutionary algorithms (EAs) are being adopted to optimize the kernel parameters. However, recently some works have concentrated on combining the SVM with meta-classifier where the kernel parameters are optimized as a single-objective using Differential Evolution. Based on this motivation, In this paper, we would like analyze the performance of the SVM with Meta-classifier by optimizing the kernel parameters through a Multi-objective approach. For the multi-objective approach, accuracy, sensitivity and specificity are employed as the objective functions. In this work, We have analyzed the performance of different Multi-objective Evolutionary algorithms (MOEAs) such as NSGAII, NSGAIII, ISDE+, PDMOEA-MR and SRA for the optimization of the kernel parameters on real-life dataset.

کلیدواژه ها:

Meta-classifier ، Multi-Objective Optimization ، Support Vector Machine ، Accuracy ، Sensitivity ، specificity.