Abstract. Damage identification and localization is fundamental in industrial engineering, since it helps perform corrective actions in time to reduce as much as possible system downtime, operational costs, perform quick maintenance and avoid failure. Recently, structural health monitoring has found in machine learning an extremely useful tool, making the monitoring of complex systems more manageable. In this work, composite plates manufactured with the purpose of damping vibrations in aerospace structures are experimentally tested; the strong damping suddenly reduces the vibrations, leading to responses very similar to one another, without noticeable or structured differences between undamaged and damaged plates. To overcome this issue, machine learning methods are applied. Decision trees-based methods are chosen since they provide a combination of feature selection capabilities and robust classification performances. The used methods are decision trees themselves and two boosting methods: AdaBoost and RUSBoost. All three methods perform well in identifying damaged plates, the type (thickness damage and debonding) and sub-type of damage (thickness/debonding of types A and B).