Unmasking ransomware attacks

1. A computer-implemented method performed by a computer system having a memory and at least one hardware processor, the computer-implemented method comprising:

generating a first prediction that a file system comprising a plurality of files has been attacked by ransomware based on snapshot metadata of the file system using a snapshot-level machine learning prediction model, the snapshot metadata comprising a plurality of file change data indicating a plurality of file change events that have been performed on the file system;

in response to the first prediction that the file system has been attacked by the ransomware using the snapshot-level machine learning prediction model, generating a corresponding classification for each one of the plurality of files in the file system based on the plurality of file change data using a file-level machine learning prediction model that is different from the snapshot-level machine learning prediction model, the corresponding classification indicating whether the corresponding one of the plurality of files has been targeted by the ransomware for encryption;

determining that one or more of the plurality of files have been targeted by the ransomware for encryption based on the corresponding classification for each one of the one or more of the plurality of files; and

causing the corresponding classification for each one of the one or more of the plurality of files to be displayed on a computing device of a user.

2. The computer-implemented method of claim 1 , wherein the generating the first prediction comprises:

obtaining a metadata file from a storage as a service data store, the metadata file comprising the snapshot metadata in Sorted Strings Table (SSTable) format; and

feeding the metadata file into the snapshot-level machine learning prediction model, the snapshot-level machine learning prediction model generating the first prediction based on the metadata file.

3. The computer-implemented method of claim 1 , wherein the generating the corresponding classification for each one of the plurality of files in the file system based on the plurality of file change data using the file-level machine learning prediction model comprises:

computing a corresponding set of feature signals for each one of the plurality of file change events based on the plurality of file change data; and

for each one of the plurality of file change events, feeding the corresponding set of feature signals into the file-level machine learning prediction model, the file-level machine learning prediction model generating the corresponding classification for each one of the plurality of files based on the corresponding set of feature signals.

4. The computer-implemented method of claim 3 , wherein the set of feature signals comprises one or more matches between corresponding file paths of the file change events and one or more file path patterns that are determined to be associated with ransomware.

5. The computer-implemented method of claim 3 , wherein the set of feature signals comprises a change in file size.

6. The computer-implemented method of claim 3 , wherein the set of feature signals comprises a change in file entropy.

7. The computer-implemented method of claim 3 , wherein the set of feature signals comprises a compression ratio statistic.

8. The computer-implemented method of claim 3 , wherein the set of feature signals comprises one or more statistics for changes to a directory of the file system.

9. A system of comprising:

at least one hardware processor; and

a non-transitory computer-readable medium storing executable instructions that, when executed, cause the at least one hardware processor to perform operations comprising:

generating a first prediction that a file system comprising a plurality of files has been attacked by ransomware based on snapshot metadata of the file system using a snapshot-level machine learning prediction model, the snapshot metadata comprising a plurality of file change data indicating a plurality of file change events that have been performed on the file system;

in response to the first prediction that the file system has been attacked by the ransomware using the snapshot-level machine learning prediction model, generating a corresponding classification for each one of the plurality of files in the file system based on the plurality of file change data using a file-level machine learning prediction model that is different from the snapshot-level machine learning prediction model, the corresponding classification indicating whether the corresponding one of the plurality of files has been targeted by the ransomware for encryption;

determining that one or more of the plurality of files have been targeted by the ransomware for encryption based on the corresponding classification for each one of the one or more of the plurality of files; and

causing the corresponding classification for each one of the one or more of the plurality of files to be displayed on a computing device of a user.

10. The system of claim 9 , wherein the generating the first prediction comprises:

obtaining a metadata file from a storage as a service data store, the metadata file comprising the snapshot metadata in Sorted Strings Table (SSTable) format; and

feeding the metadata file into the snapshot-level machine learning prediction model, the snapshot-level machine learning prediction model generating the first prediction based on the metadata file.

11. The system of claim 9 , wherein the generating the corresponding classification for each one of the plurality of files in the file system based on the plurality of file change data using the file-level machine learning prediction model comprises:

computing a corresponding set of feature signals for each one of the plurality of file change events based on the plurality of file change data; and

for each one of the plurality of file change events, feeding the corresponding set of feature signals into the file-level machine learning prediction model, the file-level machine learning prediction model generating the corresponding classification for each one of the plurality of files based on the corresponding set of feature signals.

12. The system of claim 11 , wherein the set of feature signals comprises one or more matches between corresponding file paths of the file change events and one or more file path patterns that are determined to be associated with ransomware.

13. The system of claim 11 , wherein the set of feature signals comprises a change in file size.

14. The system of claim 11 , wherein the set of feature signals comprises a change in file entropy.

15. The system of claim 11 , wherein the set of feature signals comprises a compression ratio statistic.

16. The system of claim 11 , wherein the set of feature signals comprises one or more statistics for changes to a directory of the file system.

17. A non-transitory machine-readable storage medium tangibly embodying a set of instructions that, when executed by at least one hardware processor, causes the at least one processor to perform operations comprising:

generating a first prediction that a file system comprising a plurality of files has been attacked by ransomware based on snapshot metadata of the file system using a snapshot-level machine learning prediction model, the snapshot metadata comprising a plurality of file change data indicating a plurality of file change events that have been performed on the file system;

in response to the first prediction that the file system has been attacked by the ransomware using the snapshot-level machine learning prediction model, generating a corresponding classification for each one of the plurality of files in the file system based on the plurality of file change data using a file-level machine learning prediction model that is different from the snapshot-level machine learning prediction model, the corresponding classification indicating whether the corresponding one of the plurality of files has been targeted by the ransomware for encryption;

determining that one or more of the plurality of files have been targeted by the ransomware for encryption based on the corresponding classification for each one of the one or more of the plurality of files; and

causing the corresponding classification for each one of the one or more of the plurality of files to be displayed on a computing device of a user.

18. The non-transitory machine-readable storage medium of claim 17 , wherein the generating the first prediction comprises:

obtaining a metadata file from a storage as a service data store, the metadata file comprising the snapshot metadata in Sorted Strings Table (SSTable) format; and

feeding the metadata file into the snapshot-level machine learning prediction model, the snapshot-level machine learning prediction model generating the first prediction based on the metadata file.

19. The non-transitory machine-readable storage medium of claim 17 , wherein the generating the corresponding classification for each one of the plurality of files in the file system based on the plurality of file change data using the file-level machine learning prediction model comprises:

computing a corresponding set of feature signals for each one of the plurality of file change events based on the plurality of file change data; and

for each one of the plurality of file change events, feeding the corresponding set of feature signals into the file-level machine learning prediction model, the file-level machine learning prediction model generating the corresponding classification for each one of the plurality of files based on the corresponding set of feature signals.

20. The non-transitory machine-readable storage medium of claim 19 , wherein the set of feature signals comprises one or more matches between corresponding file paths of the file change events and one or more file path patterns that are determined to be associated with ransomware, a change in file size, a change in file entropy, a compression ratio statistic, and one or more statistics for changes to a directory of the file system.