In order to advance population diversity and ergodic property for fruit fly optimization algorithm, enhance its convergence precision effectively, an algorithm named improving fruit fly optimization algorithm (abbreviated as IFOA) is proposed in this paper. The simulation experiment shows that this algorithm maintains changing in scale and balances the overall and local searching capability. In order to improve the randomness and blindness in choosing SVM model parameter artificially, enhancing accuracy for pattern classification at the same time. A method using IFOA in the field of SVM model parameter optimization is put forward and established. In this method, IFOA is applied into penalty factor and kernel function parameters optimization for SVM, with which the optimal model parameters will be chosen and the optimal SVM model can be established. This model is used in pattern classification research for wine data in UCI machine study database, different algorithms were used for comparison, the result shows that, the improved FOA has a fast speed in convergence and high efficiency in optimization, a better classification accuracy could be reached for IFOA-SVM model. The effectiveness for IFOA in wine database classification is proved thereby.