Feature Selection for Data Driven Prediction of Protein Model Quality