An Improvement On SVM Algorithm And Its Application On Time Series

Artificial intelligent technology has been paid hot attention since BP neural network was applied successfully in predicting the time series. As a new algorithm in the area of artificial intelligences, Support Vector Machine is at the point of awl sticking out through a bag for its unique advantage. This algorithm introduces the kernel function that satisfied the K-K-T condition which ensures the only answer to question of the maximum margin hyperplane and avoids the possibility of the local minimization that occurring in the process of the training of BP algorithm. Furthermore, the Support Vector and the sample number decided the complexity of the algorithm. So it avoids the disadvantage of the limitation of the sample in other algorithms. The aim of this work is to discover the influence of the feather selection to the efficiency of the model when it applied in certain areas. An improvement on the standard model is applied in the time series forecasting. The main measures of this paper are as follows:Firstly, a comparative analysis between the traditional time series analysis algorithm and SVM is given. The traditional time series analysis algorithm such as ARCH, AR is introduced. Comparing with these algorithms, the performance standards of the BP neural network have superiority over other algorithms. Since then, the artificial intelligence technology has been given a strong appeal. In the 90's of last century, a new type of learning machine, called Support Vector Machine is applied in analyzing the time series.Secondly,a detailed explanation is given to the SVM algorithm. Support Vector Machine originates from the algorithm to solving the linear clustering question and develops to solving the unlinear clustering question. When applied in certain areas, the selection of the feature, the insensitive function and the selection of the parameter are the three aspects to improve the model's efficiency. The selection of feature in common use such as ICA, PCA is given. The paper provides an algorithm of the feature selection on sensitivity of SVR that select features based on the overall key factor. Comparing with other feature selection algorithms such as PCA, ICA, tests proved it has a better performance than other algorithms.Thirdly, a simple detection on selection of the parameter is given.