Accurate classification of pregnancy risk is crucial for early detection of complications and reducing maternal and infant mortality rates. However, existing classification models still face challenges in terms of accuracy, particularly in distinguishing between high and very high-risk classes. This study aims to improve the accuracy of pregnancy risk classification models by optimizing Support Vector Machine (SVM) using Particle Swarm Optimization (PSO) algorithm. The study utilizes the "Maternal Health Risk Data" dataset available on Kaggle, which contains information on maternal health. The data is analyzed using two models: standard SVM and SVM optimized with PSO. Evaluation is conducted by comparing the accuracy, precision, recall, and F1-score of both models. The application of SVM without PSO resulted in an accuracy of 77.72%, whereas after optimization with PSO, the accuracy increased to 89.92%. Significant improvements were also observed in precision (from 79.37% to 89.49%) and recall (from 78.81% to 89.55%). The F1-score of the PSO-based SVM model reached 89.52%, demonstrating a good balance between precision and recall.