<p>The development of artificial intelligence has promoted the expansion of high-dimensional data analysis. However, in scenarios where multi-party data are difficult to centrally share, feature selection faces the challenge of balancing privacy protection and global consistency. Therefore, this study designs a novel horizontal federated feature selection framework for artificial intelligence applications, and introduces a particle swarm optimization search mechanism and a trusted third-party mechanism into the framework. This algorithm improves the efficiency of cross-participant feature selection and approaches the global optimal solution without directly sharing the original data. Experiments have shown that as the dataset size increased, the research algorithm maintained the highest level throughout the entire process, with an initial precision of about 0.84, which approached 0.98 when the data volume reached 1,000. The accuracy has increased from approximately 0.84 to 0.98, maintaining the highest overall level. As the number of participants increased, the accuracy increased from 89.7% to 92.8%, and the F1 score increased from 89.1% to 92.5%. The results indicate that the proposed horizontal federated particle swarm feature selection algorithm achieves better feature compression and classification performance in privacy-protected horizontal federated scenarios, balancing privacy, security, and efficiency. This study provides an effective approach for intelligent data mining under multi-agency collaboration.</p>