Research Of Parameter Optimization For SVM Based On Improved Grid Search And Its Application

Support Vector Machine(SVM), as a representative classification algorithm in Data Mining, is based on the statistical learning theory. SVM is used to build the optimal separating h yperplane through anal yzing the known data with real label. It is able to solve nonlinear, high dimensionality and small sample learning problems. The classification performance of SVM is closely related to the selection of parameters, so the parameters optimization of SVM has become a hot topic in machine learning.In recent years, many scholars have proposed or introduced such as Grid Searching Algorithms, Particle Swarm Optimization, Genetic Algorithm for SVM parameters optimization. However, the existing parameter optimization algorithm of SVM have a large amount of computation, especiall y in the face of the large-scale date, SVM overall operating efficiency is not high due to the SVM modeling is a slow process.To solve the above problems, this dissertation puts forward a kind of Improved SVM parameter optimization algorithm that is Half space and Variable step Grid Search(HVGS). The algorithm expands preset search step L times and in initial space for first time parameter search, then according to the result, the algorithm narrows search step M times and second time parameter search in half of initial space. repeat this a process and for the third times parameter search, and the search step reduced N times to preset step. HVGS can substantially improve the parameters of search efficiency and thus reduce the overall run time. After the preprocessing process with standardized attribute, principal component analysis and HVGS, we build the PHVSVM in MATLAB, also the accuracy and efficiency of PHVSVM is proved through experimensts.At last, PHVSVM is applied to the Public Sentiment Anal ysis, it can further validate the speed of the analysis process under the ensuring premise of sentiment anal ysis of recall and precision through its performance in practical application, so as to enhance Public Sentiment Analysis in real-time.