Predicting parallel application performance via machine learning approaches