Visualizing anomalous feature vectors based on data from healthcare records systems
Medication security and healthcare privacy analytics systems are described that enable users to search for and process stored healthcare environment data. The medication security and healthcare privacy analytics systems receive and correlate data from a plurality of data sources, including medication dispensing systems, healthcare employee records, and patient records. The medication security and healthcare privacy analytics systems generate a plurality of feature vectors from processed healthcare environment data. The visualizations are created using datasets generated by clustering algorithms and can indicate those feature vectors from the plurality of feature vectors whose data indicate anomalous interactions with various systems (e.g., indicative of unexpected or non-customary events).
1. A computer-implemented method, comprising:
generating, using one or more processors, a first set of timestamped events from first data obtained from a first database and a second set of timestamped events from second data obtained from a second database, wherein the first data corresponds to instances of interactions with patient medical records by a plurality of users and the second data corresponds to employee attendance records of the plurality of users;
filtering the first set of timestamped events and the second set of timestamped events using one or more dataset filters;
generating a plurality of feature vectors, wherein each feature vector of the plurality of feature vectors corresponds to a respective user of the plurality of users, wherein each of the plurality of feature vectors includes a plurality of features determined based on the one or more dataset filters, wherein each of the plurality of feature vectors is multi-dimensional;
using a machine learning algorithm to reduce a number of dimensions of each feature vector of the plurality of feature vectors to three dimensions; and
generating a visualization plotting the plurality of feature vectors in a three-dimensional space displayed in a graphical user interface (GUI), the visualization including a clustered representation of the plurality of feature vectors and including an indication of one or more anomalous feature vectors.
2. The method of claim 1 , further comprising:
applying a clustering algorithm to generate the clustered representation of the plurality of feature vectors;
identifying a centroid of the clustered representation of the plurality of feature vectors; and
for each feature vector of the plurality of feature vectors:
comparing a vertex corresponding to the feature vector against a threshold value, and
determining that a user corresponding to the feature vector is an anomalous user when a distance of the vertex from the centroid is greater than the threshold value.
3. The method of claim 1 , wherein the indication of the one or more anomalous feature vectors includes user information for at least one user associated with one or more anomalous feature vectors.
4. The method of claim 1 , wherein the indication of the one or more anomalous feature vectors is displayed as a first graphical representation on the visualization, wherein an indication of one or more non-anomalous feature vectors is displayed as a second graphical representation on the visualization, and wherein the first graphical representation is different from the second graphical representation.
5. The method of claim 1 , wherein the GUI includes elements enabling selection of filtering options, and wherein the filtering options include at least one of: a department, a user action precision value, a user title precision value, a timeframe resolution value, or a timeframe.
6. The method of claim 5 , wherein a number of dimensions of the plurality of feature vectors is based on a user action precision value, a timeframe resolution value, and a user title precision value.
7. The method of claim 1 , further comprising:
generating a dataset results list based on the first set of timestamped events and the second set of timestamped events, the dataset results list indicating the one or more anomalous feature vectors; and
causing a display of the dataset results list in the GUI.
8. A non-transitory computer-readable storage medium storing instructions which, when executed by one or more processors, cause performance of operations comprising:
generating, using one or more processors, a first set of timestamped events from first data obtained from a first database and a second set of timestamped events from second data obtained from a second database, wherein the first data corresponds to instances of interactions with patient medical records by a plurality of users and the second data corresponds to employee attendance records of the plurality of users;
filtering the first set of timestamped events and the second set of timestamped events using one or more dataset filters;
generating a plurality of feature vectors, wherein each feature vector of the plurality of feature vectors corresponds to a respective user of the plurality of users, wherein each of the plurality of feature vectors includes a plurality of features determined based on the one or more dataset filters, wherein each of the plurality of feature vectors is multi-dimensional;
using a machine learning algorithm to reduce a number of dimensions of each feature vector of the plurality of feature vectors to three dimensions; and
generating a visualization plotting the plurality of feature vectors in a three-dimensional space displayed in a graphical user interface (GUI), the visualization including a clustered representation of the plurality of feature vectors and including an indication of one or more anomalous feature vectors.
9. The non-transitory computer-readable storage medium of claim 8 , wherein the instructions, when executed by the one or more processors, further cause performance of operations comprising:
applying a clustering algorithm to generate the clustered representation of the plurality of feature vectors;
identifying a centroid of the clustered representation of the plurality of feature vectors; and
for each feature vector of the plurality of feature vectors:
comparing a vertex corresponding to the feature vector against a threshold value, and
determining that a user corresponding to the feature vector is an anomalous user when a distance of the vertex from the centroid is greater than the threshold value.
10. The non-transitory computer-readable storage medium of claim 8 , wherein the indication of the one or more anomalous feature vectors includes user information for at least one user associated with one or more anomalous feature vectors.
11. The non-transitory computer-readable storage medium of claim 8 , wherein the indication of the one or more anomalous feature vectors is displayed as a first graphical representation on the visualization, wherein an indication of one or more non-anomalous feature vectors is displayed as a second graphical representation on the visualization, and wherein the first graphical representation is different from the second graphical representation.
12. The non-transitory computer-readable storage medium of claim 8 , wherein the GUI includes elements enabling selection of filtering options, and wherein the filtering options include at least one of: a department, a user action precision value, a user title precision value, a timeframe resolution value, or a timeframe.
13. The non-transitory computer-readable storage medium of claim 12 , wherein a number of dimensions of the plurality of feature vectors is based on the user action precision value, the timeframe resolution value, and the user title precision value.
14. The non-transitory computer-readable storage medium of claim 12 , further comprising:
generating a dataset results list based on the first set of timestamped events and the second set of timestamped events, the dataset results list indicating the one or more anomalous feature vectors; and
causing a display of the dataset results list in the GUI.
15. An apparatus, comprising:
one or more processors; and
a non-transitory computer-readable storage medium storing instructions which, when executed by the one or more processors, causes the apparatus to:
generate a first set of timestamped events from first data obtained from a first database and a second set of timestamped events from second data obtained from a second database, wherein the first data corresponds to instances of interactions with patient medical records by a plurality of users and the second data corresponds to employee attendance records of the plurality of users;
filter the first set of timestamped events and the second set of timestamped events using one or more dataset filters;
generate a plurality of feature vectors, wherein each feature vector of the plurality of feature vectors corresponds to a respective user of the plurality of users, wherein each of the plurality of feature vectors includes a plurality of features determined based on the one or more dataset filters, wherein each of the plurality of feature vectors is multi-dimensional;
use a machine learning algorithm to reduce a number of dimensions of each feature vector of the plurality of feature vectors to three dimensions; and
generate a visualization plotting the plurality of feature vectors in a three-dimensional space displayed in a graphical user interface (GUI), the visualization including a clustered representation of the plurality of feature vectors and including an indication of one or more anomalous feature vectors.
16. The apparatus of claim 15 , wherein the instructions, when executed by the one or more processors, further causes the apparatus to:
apply a clustering algorithm to generate the clustered representation of the plurality of feature vectors;
identify a centroid of the clustered representation of the plurality of feature vectors; and
for each feature vector of the plurality of feature vectors:
compare a vertex corresponding to the feature vector against a threshold value, and
determine that a user corresponding to the feature vector is an anomalous user when a distance of the vertex from the centroid is greater than the threshold value.
17. The apparatus of claim 15 , wherein the indication of the one or more anomalous feature vectors includes user information for at least one user associated with one or more anomalous feature vectors.
18. The apparatus of claim 15 , wherein the indication of the one or more anomalous feature vectors is displayed as a first graphical representation on the visualization, wherein an indication of one or more non-anomalous feature vectors is displayed as a second graphical representation on the visualization, and wherein the first graphical representation is different from the second graphical representation.
19. The apparatus of claim 15 , wherein the GUI includes elements enabling selection of filtering options, and wherein the filtering options include at least one of: a department, a user action precision value, a user title precision value, a timeframe resolution value, or a timeframe.
20. The apparatus of claim 19 , wherein a number of dimensions of the plurality of feature vectors is based on the user action precision value, the timeframe resolution value, and the user title precision value.