A Method for Stopping Active Learning Based on Stabilizing Predictions and the Need for User-Adjustable Stopping