Study on data reduction technique for incremental training of SVDD

In this paper, a data reduction technique is proposed for incremental learning of SVDD. Two methods are used in order to train a lot of data. One method is to remove redundant data. The other one is to remove the data that almost does not affect to next training. The removed data is located inside of the boundary mostly. As a result, a computation load is reduced and the storage space can be more saved as well. The Temperature data of two Motor-Generators in commercial hybrid electric vehicle is adopted in this study. The performance characteristics in terms of computation time and the storage space are compared with conventional method. The data inside the boundary via calculation is selected to be removed in the proposed technique. One of the big advantages in the proposed techniques is that the same result is hold even using the set of reduced data. Simulation confirms, in addition, that the proposed data reduction technique exhibits acceptable error without losing its performance.