Table 3 The evaluation metrics
From: Towards a deep learning-based outlier detection approach in the context of streaming data
Evaluation metric | Definition | Equation |
|---|---|---|
Accuracy | The percentage of accurately predicted data to the total amount of data | \(\frac{TP+TN}{TP+FP+FN+TN}\) |
True Negative Rate (Specificity) | The proportion of correctly predicted inlier data to total normal data | \(\frac{TN}{FP+TN}\) |
False alarm rate | The proportion of outliers that were incorrectly predicted in comparison to all inlier instances | \(\frac{FP}{FP+TN}\) |
False negative rate (Miss rate) | The proportion of incorrectly predicted inlier data to total outliers It indicates the probability that the model will miss the outlier cases | \(\frac{FN}{TP+FN}\) |
Precision | The proportion of correctly predicted outliers to all predicted outlier instances | \(\frac{TP}{TP+FP}\) |
Recall (Sensitivity or hit rate) | The proportion of correctly predicted outliers to all actual outliers | \(\frac{TP}{TP+FN}\) |
F1-score | The harmonic mean of precision and recall values | \(F= 2\left(\frac{Precision*Recall}{recision+Recall}\right)\) |