Skip to main content
Fig. 5 | BMC Medical Research Methodology

Fig. 5

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

Fig. 5

FindFPOF anomaly ranking over autoencoder anomaly ranking for the 157 different implausible records in the FindFPOF, autoencoder, and random samples. Each implausible record is ranked based on the FindFPOF anomaly score and the autoencoder anomaly score. High ranks correspond to anomalous records and low ranks correspond to normal records. The autoencoder finds the implausibilities identified by FindFPOF earlier than the other way around. All implausible records of the FindFPOF sample are in the 4477 (\(21\%\) of all 21,104 records) highest-ranked records of the autoencoder. The implausible records of the autoencoder sample are in the 9084 (\(43\%\) of all 21,104 records) highest-ranked records of FindFPOF. Note that the scales are square root transformed

Back to article page

BMC Medical Research Methodology

ISSN: 1471-2288

Contact us