LP SVM with A Novel Similarity function Outperforms Powerful LP-QP-Kernel-SVM Considering Efficient Classification

– While the core quality of SVM comes from its ability to get global optima, classification performance also depends on computing kernels. However, while this kernel-complexity generates power machine, it is responsible for compu- tational load execute kernel. Moreover, insisting a similarity function be positive definite kernel demands some properties satisfied that seem unproductive sometimes raising question about which measures used classifier. We model Vapnik’s LPSVM proposing new replacing func- tion. Following strategy ”Accuracy first, speed second”, we have modelled mathematically well-defined depending analysis as well geometry and complex enough train machine generating solid generalization ability. Being consistent with theory learning by Balcan Blum [1], our similar ity does not need valid less computational cost executing compared counterpart like RBF or other kernels provides sufficient classifier using optimal complexity. Benchmarking shows based poses test error 0.86 times most powerful QP but only 0.40 cost.