Uncovering the predominant factors influencing the transient stability margin in power systems with increasing renewable generation using interpretable machine learning