Evaluation Methodology for Attacks Against Confidence Thresholding Models