Natural frequency and mode shape differences play a vital role to distinguish between healthy and unhealthy structures. As per the proposed method, the unhealthy structure leads to crack identification in terms of the location and its severity. There are several methods defined in early stage of the research work, but multi-output least square support vector regression machine (MLS-SVR) is an easiest and quick tool in this field of application. MLS-SVR by its tendency can be used for data regression, which can be used to train the training data and test the testing data. The prospective effectiveness of the multi-output regression machine is to map two or more variables of input feature space to two or more variables of output feature space. In our research work, the natural frequencies of the beam structure are considered as input feature space and the crack depth and location are treated as output feature space. The research work initiated from theoretical approach and ended with a comparative statement with MLS-SVR, and it is found that MLS-SVR is one of the best methods for crack detection.