Parameter Optimization is the most crucial task during any model development. The task becomes challenging when the model involves multiple parameters with interrelationships among them. This work concentrates on parameter optimization for Support Vector Regression of atmospheric variables. Support vector regression involves the parameter C which controls the smoothness of the approximating function and epsilon that determines the error tolerance margin. Due to the favorable performance of Radial Basis Function kernel in the previous studies on atmospheric variable regression it is adopted in this study. Hence we require to optimize the parameter gamma (_) in addition to the support vector regression parameters C and epsilon.
[...] “Support vector regression for realtime flood stage forecasting.” Journal of Hydrology pp 704- Ronan Collobert and Samy Benegio Torch: Support Vector Machines for LargeScale Regression Problems.” Journal of Machine Learning Research pp Smola A.J, and Scholkopf Tutorial on support vector regression,” Neuro COLT Technical Report NC-TR-98-030, Royal Holloway College, Uni of London, UK Stanislaw Osowski and Konrad Garanty, “Forecasting of daily meteorological pollution using wavelets and support vector machine.” Engineering Applications of Artificial Intelligence pp 745- Wei-Zhen Lu. Wen-Jian Wang. [...]
[...] The proposed algorithm is applied for parameter optimization of support vector regression and the results at various stages are tabulated and shown in figs and 4. Fig 2 shows the variation of Mean Square Error with C and the values are tabulated in table 1. It can be observed that for C = the model gives the best performance with minimal MSE. Figure 3 shows the variation of MSE with whose values are tabulated in table 2. It can be observed that has a significant effect on the performance of the system. [...]
[...] Since it is not known in advance which values of the parameters are the best for a problem, some kind of parameter optimization technique must be adopted PROPOSED APPROACH FOR PARAMETER OPTIMIZATION In this work support vector regression is used for predicting the maximum temperature of a day based on the maximum temperature of previous n days where n is the optimal length of the span. The value of n referred to as order of the model is found by experimentation. [...]
[...] The proposed recursive algorithm is used to perform the fine grain search in the optimal region during the second stage of the grid search method. This recursive algorithm is an intelligent method that performs the fine grain search by eliminating the exhaustive micro level search in the optimal region. The algorithm is described below. P contains the default values of the parameters. For the first call P contains values of the parameters returned by the coarse analysis of grid search. [...]
[...] We have devised a new parameter optimization algorithm to explore multidimensional parameter space at microlevel. Support Vector Regression requires the setting of parameters along with the kernel specific parameters like γ for RBF kernel. This paper presents the proposed algorithm and uses it for SVR parameter setting. The proposed algorithm is tested on weather dataset to predict the maximum temperature of a day based on the data of the previous n days [ 6]. The paper is structured as follows. [...]
using our reader.