We present the first automated approach, named Casboundary, to define cassette boundaries, using different machine learning approaches. The Casboundary can obtain high predicted performances for cassette boundaries detection with an average of 0.86 in set similarity and 1.09 in boundary loss measure