Static anomaly-based detection of malware files
A protection application detects and remediates malicious files on a client. The protection application trains models using known samples of static clean files, and the models characterize features of the clean files. A model may be selected based on metadata obtained from a target file. By processing features of the clean files and features of the target file, the model may generate an anomaly score indicating a level of dissimilarity between the target file and the sample. The protection application compares the anomaly score to one or more threshold scores to classify the target file. Additionally, the target file may be provided to a security server to check against a whitelist or blacklist for classification. Responsive to a classification as malicious, the protection application remediates the target file on the client.
1. A method for detecting anomalous files, the method comprising:
determining a plurality of subclasses of a plurality of files on a client;
classifying a file of the plurality of files as belonging to a subclass of the plurality of subclasses;
selecting a machine learning model derived from a training set that includes only clean files belonging to the subclass of the plurality of subclasses without the training set including malicious files;
generating, by a processor, an anomaly score of the file by applying the selected machine learning model to the file, the anomaly score indicating a level of dissimilarity between features of the file and a plurality of features of the training set that includes only clean files, wherein generating the anomaly score comprises:
determining a first distance indicative of dissimilarity between a first feature of the features of the file and a second feature of the plurality of features of the training set,
determining a second distance indicative of dissimilarity between a third feature of the features of the file and a fourth feature of the plurality of features of the training set, and
combining the first distance and the second distance to generate the anomaly score:
classifying the file as anomalous based on the anomaly score; and
remediating the file by the client responsive to the classification of the file.
2. The method of claim 1 , further comprising:
determining a mean feature vector of the plurality of features of the training set that includes only clean files; and
wherein the anomaly score is generated by determining distances between the features of the file and the mean feature vector.
3. The method of claim 1 , further comprising:
receiving, at the client from a security server, a plurality of machine learning models each associated with a different subclass, the machine learning model selected from the plurality of machine learning models.
4. The method of claim 1 , wherein classifying the file as anomalous comprises:
responsive to comparing the anomaly score against a threshold score:
providing the file to a security server for comparison against a blacklist or a whitelist; and
classifying the file as anomalous responsive to receiving an indication from the security server that the file is on the blacklist or not on the whitelist.
5. The method of claim 1 , further comprising:
providing a notification to a user of the client responsive to classifying the file as anomalous, the notification indicating a risk severity level proportional to the anomaly score.
6. The method of claim 1 , wherein remediating the file by the client comprises:
providing, to a user of the client, an option to remove the file from the client.
7. The method of claim 6 , further comprising:
responsive to receiving a selection from the user to retain the file, providing feedback to a security server for retraining the selected machine learning model, the selection responsive to the provided option to remove the file from the client.
8. A non-transitory computer-readable storage medium storing instructions for detecting anomalous files, the instructions when executed by a processor causing the processor to perform steps including:
determining a plurality of subclasses of a plurality of files on a client;
classifying a file of the plurality of files as belonging to a subclass of the plurality of subclasses;
selecting a machine learning model derived from a training set that includes only clean files belonging to the subclass of the plurality of subclasses without the training set including malicious files;
generating, by a processor, an anomaly score of the file by applying the selected machine learning model to the file, the anomaly score indicating a level of dissimilarity between features of the file and a plurality of features of the training set that includes only clean files, wherein generating the anomaly score comprises:
determining a first distance indicative of dissimilarity between a first feature of the features of the file and a second feature of the plurality of features of the training set,
determining a second distance indicative of dissimilarity between a third feature of the features of the file and a fourth feature of the plurality of features of the training set, and
combining the first distance and the second distance to generate the anomaly score;
classifying the file as anomalous based on the anomaly score; and
remediating the file by the client responsive to the classification of the file.
9. The non-transitory computer-readable storage medium of claim 8 , comprising further instructions when executed by the processor causing the processor to perform steps including:
determining a mean feature vector of the plurality of features of the training set that includes only clean files; and
wherein the anomaly score is generated by determining distances between the features of the file and the mean feature vector.
10. The non-transitory computer-readable storage medium of claim 8 , comprising further instructions when executed by the processor causing the processor to perform steps including:
receiving, at the client from a security server, a plurality of machine learning models each associated with a different subclass, the machine learning model selected from the plurality of machine learning models.
11. The non-transitory computer-readable storage medium of claim 8 , wherein classifying the file as anomalous comprises:
responsive to comparing the anomaly score against a threshold score:
providing the file to a security server for comparison against a blacklist or a whitelist; and
classifying the file as anomalous responsive to receiving an indication from the security server that the file is on the blacklist or not on the whitelist.
12. The non-transitory computer-readable storage medium of claim 8 , comprising further instructions when executed by the processor causing the processor to perform steps including:
providing a notification to a user of the client responsive to classifying the file as anomalous, the notification indicating a risk severity level proportional to the anomaly score.
13. The non-transitory computer-readable storage medium of claim 8 , wherein remediating the file by the client comprises:
providing, to a user of the client, an option to remove the file from the client.
14. The non-transitory computer-readable storage medium of claim 13 , comprising further instructions when executed by the processor causing the processor to perform steps including:
responsive to receiving a selection from the user to retain the file, providing feedback to a security server for retraining the selected machine learning model, the selection responsive to the provided option to remove the file from the client.
15. A computing system comprising:
a processor; and
a non-transitory computer-readable storage medium storing instructions for detecting anomalous files, the instructions when executed by the processor causing the processor to perform steps including:
determining a plurality of subclasses of a plurality of files on a client;
classifying a file of the plurality of files as belonging to a subclass of the plurality of subclasses;
selecting a machine learning model derived from a training set that includes only clean files belonging to the subclass of the plurality of subclasses without the training set including malicious files;
generating, by a processor, an anomaly score of the file by applying the selected machine learning model to the file, the anomaly score indicating a level of dissimilarity between features of the file and a plurality of features of the training set that includes only clean files, wherein generating the anomaly score comprises:
determining a first distance indicative of dissimilarity between a first feature of the features of the file and a second feature of the plurality of features of the training set,
determining a second distance indicative of dissimilarity between a third feature of the features of the file and a fourth feature of the plurality of features of the training set, and
combining the first distance and the second distance to generate the anomaly score;
classifying the file as anomalous based on the anomaly score; and
remediating the file by the client responsive to the classification of the file.
16. The system of claim 15 , wherein the non-transitory computer-readable storage medium stores further instructions when executed by the processor causing the processor to perform steps including:
determining a mean feature vector of the plurality of features of the training set that includes only clean files; and
wherein the anomaly score is generated by determining distances between the features of the file and the mean feature vector.
17. The system of claim 15 , wherein the non-transitory computer-readable storage medium stores further instructions when executed by the processor causing the processor to perform steps including:
receiving, at the client from a security server, a plurality of machine learning models each associated with a different subclass, the machine learning model selected from the plurality of machine learning models.
18. The system of claim 15 , wherein classifying the file as anomalous comprises:
responsive to comparing the anomaly score against a threshold score:
providing the file to a security server for comparison against a blacklist or a whitelist; and
classifying the file as anomalous responsive to receiving an indication from the security server that the file is on the blacklist or not on the whitelist.
19. The system of claim 15 , wherein remediating the file by the client comprises:
providing, to a user of the client, an option to remove the file from the client.
20. The system of claim 19 , wherein the non-transitory computer-readable storage medium stores further instructions when executed by the processor causing the processor to perform steps including:
responsive to receiving a selection from the user to retain the file, providing feedback to a security server for retraining the selected machine learning model, the selection responsive to the provided option to remove the file from the client.