Skip to main content
Fig. 1 | BMC Medical Research Methodology

Fig. 1

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

Fig. 1

Selection and evaluation approach of tumor records. We separately rank all 21,104 records with unknown plausibility by the FindFPOF and autoencoder anomaly scores. From the 21,104 ranked records, we select the 300 highest-ranked (most anomalous) records according to either FindFPOF or the autoencoder. Additionally, we randomly select 300 records as a baseline. Out of a total of 900 selected records, 785 were different. Domain experts evaluated the plausibility of these 785 different records. The 900 (785 different) evaluated records with known plausibility were analyzed for this study

Back to article page