Aiming at the shortcomings of support vector machines in wrapper feature selection, such as poor classification effect, redundant subset selection, and computational performance that are easily affected by kernel function parameters, the meta-heuristic optimization algorithm is used to optimize it simultaneously. Firstly, the local search ability and the exploration and utilization solution space ability of the bald eagle search algorithm are improved by using the Levy flight strategy and simulated annealing mechanism, the test results of the standard function prove that the improvement is effective. Then, the kernel function parameters of the support vector machine are taken as the optimization objective, and the improved algorithm is used to search for the optimal kernel function parameters in the wrapper feature selection model and it obtains the corresponding feature subset simultaneously. Finally, a feature selection simulation is performed on the 12 standard data sets of the UCI repository, and the average classification accuracy, the number of selected features and the fitness value are comprehensively evaluated and analyzed. The experimental results show that the proposed algorithm can effectively reduce the feature dimension and achieve data classification more accurately. Compared with the original algorithm and other nonlinear optimization algorithms, it has certain engineering application value.