Abstract: Files of computer documents are classified into confidential levels without extracting and analyzing contents of the files. Files created by particular users may be clustered into groups of files based on file characteristics, such as file type (e.g., file extension) and file naming convention. A prediction confidential score may be generated for each group of files. A log of a file retention resource may be consulted to identify files created by users. A file created by a user may be assigned a prediction confidential score of a group of files having the same file characteristic as the file and created by the same user. The prediction confidential score may be used to determine a confidential level of the file when the file is found to be inaccessible.