Optimization of Support Vector Machine Model Based on Kernel Function Selection and Its Application in Electromagnetic Environment Adaptability Analysis

Addressing the challenge of evaluating the adaptability of electronic equipment in complex electromagnetic environments with small sample sizes, this paper focuses on the Support Vector Machine (SVM) method, emphasizing the impact of kernel function selection on model performance. A kernel calibration strategy based on inter-class and intra-class distances is proposed. Experimental results show that using the Radial Basis Function (RBF) kernel combined with the kernel calibration strategy improves the overall prediction accuracy to 82.3%, a 6.7% increase over the original model. The prediction accuracy for intermediate adaptability levels (second to fourth classes) improves by $\mathbf{1 2. 2 \%}$ to $\mathbf{2 0. 3 \%}$, validating the kernel function’s role in enhancing intra-class compactness. Meanwhile, the prediction reliability for extreme adaptability levels (first and fifth classes) remains stable ($\gt90 \%$). This study provides theoretical basis and practical guidance for kernel function selection in electromagnetic environment data analysis.