Vibration-induced ride discomfort impacts vehicle performance and operator well-being, necessitating accurate and interpretable predictive models. This study presents an Interpretable multi-model machine learning (ML) framework combining six ensemble models: Random Forest, Extra Trees, Bagging, Gradient Boosting, Extreme Gradient Boosting, and Adaptive Boosting Regressors. The Meta-Learner integrates a Multilayer Perceptron (MLP) and Random Forest Regressor (RFR) to enhance predictive performance. In the results, the Bayesian-optimized Gradient Boosting model, found as the best among the individual models, achieved a Mean Squared Error (MSE) of 0.1%, a Root Mean Squared Error (RMSE) of 3.8%, a Coefficient of Determination (R2) of 94.8%, and a Mean Absolute Error (MAE) of 2.8%. However, the RFR resulted in the same MSE but demonstrated superior performance with a lower RMSE of 2.3%, an improved R2 of 98.0%, and a reduced MAE of 1.7%. While the MLP model was competitive compared to individual models, it had higher mean prediction errors than the RFR. These results highlight the Meta-Learner's effectiveness in minimizing prediction errors and capturing complex data relationships. Interpretability analyses using Shapley Additive Explanations and Local Interpretable Model-agnostic Explanations identified speed-related features as the most influential predictors, while the Bagging Regressor provided significant contributions to the Meta-Learner's performance. In conclusion, this study establishes a ML framework that improves predictive performance and ensures model transparency. The findings advance comfort assessment methodologies and support the development of human-centered vehicle designs for vibration-sensitive applications.