Vibration-based Structural Health Monitoring (VSHM) is becoming one of the most commonly used methods for damage diagnosis and long term monitoring. In data-driven VSHM methods, Damage Sensitive Features (DSFs) extracted from vibration responses are compared with reference models of the healthy state for long-term monitoring and damage identification of the structure of interest. However, data-driven VSHM faces a crucial problem - the DSFs are not only sensitive to damage but also to Environmental and Operational Variabilities (EOV). Machine learning and related methods, enabled by the availability of large monitoring datasets, can be used for mitigation of EOV in DSFs. EOV mitigation methods can be grouped into implicit and explicit methods. In the former, EOVs are compensated solely on the basis of the patterns identified in DSFs in the reference state. While the latter utilize measurements of the EOVs in addition to DSFs to build a cause-effect model, typically in the form of a regression. In this chapter, these two methods are discussed and illustrated via two different approaches: an artificial neural network for metric learning (implicit) and a multivariate nonlinear regression (explicit). The rationale and limitations of both methods are studied on an operating wind turbine where different damage scenarios were introduced. Additionally, the best practices of each procedure are presented through a comprehensive discussion of their potential advantages and drawbacks.