Challenges and limitations of synthetic minority oversampling techniques in machine learning | Synapse