The Naive Bayes classifier, with its solid foundation in probability theory, exhibits significant classification advantages when dealing with datasets containing uncertain features and noise interference. With the increasing complexity of social data, the method of measuring prior probability based on proportion has to some extent limited the performance of Naive Bayes classifiers. The construction of prior probabilities is an important issue in Bayesian classification research and a crucial factor in determining the accuracy of Naive Bayes classification. How to effectively estimate and construct the highest priority probability has gradually become a research topic of concern for scholars. Therefore, this article introduces t distribution variation and adaptive weights to improve the individual update formula of Hippopotamus Optimization Algorithm (HOA). Based on this improvement, an optimization approach is proposed, which integrates the optimized algorithm with the construction of Bayesian optimal priors using both training and testing samples. This method has achieved superior classification performance. The specific process is to collect system data and divide it into training set, validation set, and test set. The model parameters obtained from the training set are used as the initial input for the Bayesian classifier in the validation set. Then, with classification accuracy as the objective function, the improved Hippopotamus Optimization Algorithm is used to search for the Bayesian optimal prior in the validation set. Finally, the optimization result is used as the prior in the test set to obtain the classification accuracy. The proposed method was tested by switching circuit system simulations and compared with other mainstream classification algorithms. The results showed that the proposed method exhibited high classification accuracy.