TWIN SUPPORT VECTOR REGRESSION BASED ON FRUIT FLY OPTIMIZATION ALGORITHM

摘要

Recently, Twin Support Vector Regression (TSVR), which determines a pair of epsilon-insensitive lower and upper bound functions by solving two related SVR-type problems, has become a new hot topic in machine learning field. However, at least four parameters should be appropriately specified in TSVR. In this paper, in order to obtain the optimal parameters of TSVR, we proposed a twin support vector regression based on fruit fly optimization algorithm. First, we represented the parameters to be optimized in TSVR by the locations of the fruit flies. Then, we used fitting regression precision as fitness function, and let fruit flies fly randomly to avoid trapping into local minimum. Finally, we could find the highest regression accuracy corresponding to the final position of the fruit flies within finite iterations. The experimental results on benchmark datasets and glutamic acid fed-batch fermentation process show that the proposed algorithm can be used to find suitable parameters for TSVR. Furthermore, our algorithm costs less optimization time than other state-of-the-art algorithms.