Skip to main content
Fig. 3 | BMC Medical Research Methodology

Fig. 3

From: Unsupervised anomaly detection of implausible electronic health records: a real-world evaluation in cancer registries

Fig. 3

Performance of both anomaly detection methods on the labeled random sample. The top row shows the sensitivity (A) and specificity (B) over the number of selected highest-ranked (most anomalous) records in the random sample. We selected the most anomalous records in the random sample according to the anomaly score of each method. The bottom row shows the ROC curve (C) and the precision-sensitivity curve (D). For more than approximately 25 selected records, FindFPOF has a higher sensitivity than the autoencoder. The specificity of both methods decreases similarly

Back to article page