Improving Logging Prediction on Imbalanced Datasets