Structural Health Monitoring (SHM) of bridges is fundamental to ensure public safety and extend the life of infrastructures. Early detection of structural damage, through the extraction of Damage Sensitive Features (DSFs), plays a key role in improving maintenance strategies and reduce the risk of structural failure. This study proposes a data-driven methodology for identifying and extracting wavelet-based DSFs, which captures critical information about the structural integrity of bridges. Starting from data recorded by the sensor networks installed on the bridge, the approach employs time-frequency analysis techniques to process and analyze large datasets, thereby enabling the extraction of features highly sensitive to damage. By leveraging wavelet decomposition, this method not only isolates features indicative of damage but also identifies the specific frequency bands where damage has the most substantial impact. A modal analysis-driven selection of wavelet decomposition levels further ensures that the analysis focuses on the most relevant frequency scales for detecting structural changes, enhancing the sensitivity and accuracy of DSF extraction. The methodology is validated using real-world data to assess the ability of the extracted DSFs to discriminate between healthy and damaged states of the bridge. Results demonstrate that the proposed approach effectively identifies early indicators of structural damage, providing valuable insights into the bridge's condition. The use of a data-driven algorithm is particularly advantageous due to its versatility and generalizability, as it is not limited to specific types of structures or conditions. This makes the approach suitable for automating the management of bridge monitoring and maintenance. Therefore, the findings suggest that time-frequency-based DSF extraction is a reliable tool for SHM systems, with significant implications for the future of bridge safety, maintenance, and infrastructure management.