Nesting Differential Evolution to Optimize the Parameters of Support Vector Machine for Gender Classification of Facial Images
Pin Liao, Sensen Wang, Xin Zhang, Kunlun Li, Mingyan Wang · Advances in intelligent systems research/Advances in Intelligent Systems Research · 2015
Support Vector Machine (SVM) is an influential and fashionable statistical learning technique for binary classification and regression. The generalization performance of SVM highly depends on proper tuning of the penalty parameter and kernel function parameter(s). An originaltechnique is proposed to quickly search the optimal SVM parameters separately by nesting two differential evolution (DE) algorithms, which can avoid repetitious costly computation and then shrink the computation cost by orders of magnitude compared to the existing approaches which tune all the parameters concurrently.The experimental results on gender classification of facial images illustrate that the proposed technique can efficiently construct an SVM classifier with significant generalization capabilities.