Reliable prediction of the seismic response of bridges is essential for ensuring structural safety and guiding effective risk mitigation strategies. While recent advancements highlight the promise of Machine Learning (ML) as an alternative to computationally intensive nonlinear time history analyses (NTHA), most existing studies are limited to specific bridge types or narrow sets of ML algorithms. Moreover, the influence of dataset size and problem complexity on predictive performance remains underexplored.

This work presents a comparative study of multiple ML techniques for estimating the seismic response of a bridge, addressing both linear and nonlinear structural behaviors. The algorithms considered include Linear Regression, Random Forest, XGBoost, Gaussian Process Regression, Multi-Layer Perceptron (MLP) networks, and Support Vector Regression (SVR). To evaluate the sensitivity of model performance to dataset size, training datasets range from 100 to 1000 samples.

Model accuracy and generalization capabilities are assessed using several statistical metrics: Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Coefficient of Determination (R²), and Mean Absolute Percentage Error (MAPE). In addition, the study examines training time across different dataset sizes to evaluate the computational efficiency of each method.

The goal is to identify the most suitable ML approaches for seismic response prediction, depending on the structural model complexity and the amount of data available. The findings offer valuable insights into balancing accuracy, computational demand, and robustness in the context of bridge seismic assessment.