Autoencoder and clustering based anomaly detection

Initially developed as a project with DHS Science and Technology, this project took place from 2018 to 2019 (pre-COVID). The approach here is to find potentially anomalous cases which are also related. To create anomaly “scores”, a neural network autoencoder is used to process over patient data from the emergency department visit and quantify how “rare” this kind of visit might be compared to all other previous visits. Then a density-based clustering is used to identify any potential clusters of anomalous cases since one single case may not warrant concern, but a group of patients with similar labs, signs/symptoms, etc might suggest common exposures and conditions.

Additional Resource Information:

  • —

  • Not currently published.

  • —
  • —
  • —

  • —
  • —
  • Decision Support tool
  • —

*Fields with “—” indicate the center did not provide information at the time of the survey. Please see the resource URL or contact the center if you would like updated information on those fields.

Center:
Resource Types:

Contact:

ForeSITE

Email:

cfa_stedi@lists.utah.edu

Organization:

Data Dictionary:

Badges: