A Novel Perspective towards SVM Combined with Autoencoder

Abstract

In this paper, we propose a novel perspective towards the hybrid algorithm about support vector machine combined with neural network. We suggest that the depth of convolution neural network is supposed to insight the view of machines to acquiring an equal level of features as human do. The kernel function of support vector machine can be grasped flexibly where the neural network makes an efficient cross calculation for features exactly instead of the kernel function but more adjustable. To develop such a coincident format, we build a hybrid model with the half former part of autoencoder working as the kernel function and support vector machine working as the core classifier, with certain ways to train the hybrid model: discrete, continuous and prejudice. The hybrid model inherits asset of each algorithm, and that process is generally subject to the objective perspective. We take the hybrid model to Covid 19 detection compared with other well-performed models, and experimental results illustrate that our perspective is advisable which achieves a state-of-the-art performance in medical scheme.