In response to the problems of low optimization accuracy, slow convergence speed, and susceptibility to local optima in the training of Multi Layer Perceptron (MLP) using traditional Butterfly Optimization Algorithm (BOA), this paper proposes a Multi strategy Hybrid Improved BOA (MSHBOA). Firstly, the initialization strategy of the optimal point set is used to enhance the uniformity of population distribution, combined with quantum search mechanism to improve local development efficiency; Secondly, the adaptive inertia weight is introduced to dynamically balance the ratio of global search and local development, and the Cauchy mutation operator is integrated to enhance the algorithm's ability to jump out of local optima; Finally, the individual perturbation strategy based on improved chaotic Tent mapping further enhances population diversity. To verify the effectiveness of MSHBOA, it was applied to the weight optimization process of MLP classifier by adaptively updating network parameters to reduce classification errors. The experiment selected the UCI standard dataset for testing, and the results showed that compared with the basic BOA and its mainstream improvement algorithms, MSHBOA improved classification accuracy by an average of 8.3%–14.7%, convergence speed by about 2.1 times, and significantly reduced the value of the cross entropy loss function. This article provides an efficient hybrid strategy framework for solving high-dimensional complex optimization problems, which has practical application value in the field of pattern recognition.