The analysis and comparison of spatial invariants shaping the structural system response is the basis of several damage identification methods, such as those employing mode shape analysis. Approaches based on strain energy assume that damage-induced changes in physical properties are more detectable if curvatures are considered. The application of physics-based techniques requiring a baseline typically provides an index distribution over the considered structure which detects local damage depending on exceeding a certain threshold (damage condition). This threshold is typically decided by setting the confidence level in assessing the damage, a key parameter for data affected by uncertainties which tunes the balance of false positives and false negatives. To increase the accuracy of the positive predictions without significantly compromising the sensitivity, an alternative is provided by the definition of a macro-index (Dessi et al., JSV 2025), which is a combination of different indices based on an ensemble learning principle. Here, four different indices are applied to experimentally identify a local reduction in the thickness of a hollow beam. It is shown that damage identification results are improved if ensemble learning is employed. Thus, different voting schemes are used for the definition of the macro-index, resulting in a comparison of their effectiveness in terms of identification accuracy.