AUC

AUC值域为0.5到1.0，0.5代表随机猜测、1.0代表完美预测，不受样本类别不平衡影响

According to common industry standards, AUC > 0.9 is excellent, 0.8–0.9 is good, 0.7–0.8 is fair, 0.5–0.7 is poor, and 0.5 is equivalent to random guessing.

AUC equals the probability that, when one positive sample and one negative sample are randomly drawn from all positive and negative samples respectively, the predicted score of the positive sample is higher than that of the negative sample.

An AUC=1 on the ROC curve indicates a perfect classifier, while AUC=0.5 is equivalent to random guessing.

AUC can be obtained by numerically integrating the area under the ROC curve using the trapezoidal rule, or equivalently computed via the Mann-Whitney U statistic.

The value of AUC ranges from 0 to 1, and a higher value indicates stronger classification ability of the model.

识别过拟合的经典方法是比较训练集AUC和验证集AUC——如果训练集AUC接近1.0而验证集AUC明显偏低，就是过拟合的典型信号

AUC（Area Under Curve）是衡量分类模型好坏最常用的指标之一

利用正样本秩次之和快速计算AUC的算法时间复杂度为O(n log n)

A random forest using only calendar features achieved a next-day wet-day AUC of 0.889 under random split, which dropped to 0.835 when transferred to future years, below the climatology baseline's 0.859.

A simple station-month climatology baseline achieved 88.4% accuracy and an AUC of 0.950 in classifying high-rainfall months.

After transfer that meets the target limit, the threshold rejects most correct detections; under rainy conditions, the area under the receiver operating characteristic curve (AUC) does not decrease.

The AUC for distinguishing requirement items from non-requirement items using wording text alone is 0.664, defining the performance ceiling achievable without relying on formatting cues.

AUC measures the probability that a model ranks a same-person pair ahead of a different-person pair among all same-person/different-person comparison pairs, takes values from 0 to 1, is better the closer to 1, and is independent of a specific threshold.

The VTF900 study shows that high AUC under small-scale galleries does not translate into reliable performance at deployment scale, and model rankings derived from hundred-person-scale galleries may undergo substantial reordering at the 900-person scale.

The DeLong test is a non-parametric statistical method used to compare whether the difference between the areas under two correlated ROC curves (AUC) is significant, taking into account the correlation of the predictions of two models on the same dataset.

AUC measures ranking ability, KS focuses on discrimination, and GINI is linearly related to AUC.

In rare-positive scenarios (such as fraud, abuse, and defect detection), AUC may look high, but because the negative pool is enormous, even a low false-positive rate produces many false alarms, resulting in very low precision.

For a binary classification feature, one should choose a threshold on a labeled holdout set reflecting the production distribution's skew, based on acceptable false-positive and false-negative costs, report F1 at that threshold, and keep AUC on the scored samples for comparing models before locking in the cutoff.

The z-score is the output statistic of the SynthID detector, measuring how far the bias of watermark tokens in the tested text deviates relative to unbiased random text; AUC measures the detector's overall ability to distinguish watermarked text from clean text, with 0.5 indicating no discrimination at all and 1.0 indicating perfect discrimination.