Structural Health Monitoring (SHM) is essential for early damage detection in structures, enabling timely maintenance and ensuring safety. Conventional single-model structural system identification (St-Id) methods are often computationally expensive and prone to ill-posedness in high-dimensional scenarios. This study explores Multi-Layer Perceptron Artificial Neural Networks (MLP ANNs), pre-trained on Finite Element (FE) model simulation data, to address these challenges and approach damage assessment as a model class selection problem. The MLP ANNs capture complex input-output relationships with multi-layer architectures, efficiently handling high-dimensional data while preserving critical information. The approach is tested on a numerical model of a representative bridge span under two damage scenarios: stiffness reduction and tendon prestress losses. The study focuses on selecting the optimal model class based on the ANN’s predictive performance, which shows high accuracy and reliability. Diverse data sources, including modal displacements, frequencies, and static rotations, are used to create robust training, validation, and testing datasets. The results confirm the ANN-based approach’s accuracy and emphasize the benefits of multi-feature approaches and data fusion in achieving reliable damage identification.