A transfer learning-driven aerodynamic noise spatial radiation characteristics prediction method was proposed to address the challenge of modeling the spatial radiation characteristics of compressor aerodynamic noise under small-sample conditions. A hybrid computational aeroacoustics (CAA) approach coupling Computational Fluid Dynamics (CFD) with the Boundary Element Method (BEM) was adopted to perform multi-condition aerodynamic noise numerical simulation for Compressor A. Simultaneously, multi-measurement-point experimental data were collected to complete accuracy verification of the numerical simulation model. Based on extensive simulation data of the target domain under a specific operating condition for Compressor B, a random forest-based pre-trained model was constructed and trained to extract common features of aerodynamic noise spatial radiation. Using a small amount of labeled data from the target domain under a specific operating condition of Compressor A, feature transfer and parameter fine-tuning were implemented on the pre-trained model to establish a transfer learning model adapted to Compressor A, enabling accurate prediction of aerodynamic noise radiation characteristics. Comparative verification indicated that the proposed transfer learning model exhibited significantly superior prediction accuracy and stability compared to traditional machine learning methods. The prediction error of the overall sound pressure level for most measurement points was controlled within 3 dB, and the model demonstrated excellent robustness and generalization ability under small-sample scenarios, thus providing reliable technical support for the rapid prediction of aerodynamic noise in different types of compressors.