Abstract: Developing accurate models to predict the vibration transmission to the seated human body is essential for studying the riding discomfort. Support Vector Regression (SVR) offers better generalization than traditional dynamic models while investigating the high-dimensional, small-sample data, although its performance depends on the hyperparameter tuning. In this study, anthropometric parameters, seat conditions and the fore-and-aft excitation amplitudes were set as input features to evaluate SVR models for predicting the seat transmissibility. Three optimization algorithms, namely Grid Search (GS), Particle Swarm Optimization (PSO), and Golden Sine Algorithm (Gold-SA), were utilized for the hyperparameter tuning of the SVR model. Results show that resonant frequencies of the fore-and-aft horizontal seat transmissibility to both the seat cushion and the backrest decreases with the increase of the horizontal excitation. Compared to the GS-SVR, the PSO-SVR improved the model’s R² value to 0.93. Furthermore, the Gold-SA-SVR achieved even higher R² values of 0.965 and 0.964 for the transmissibility to the backrest and cushion, respectively.