File Classification Containing Secret Key in Git Repository Using Character Level CNN and Shannon Entropy

Current technological developments are directly proportional to the number of information technology projects made. This project is usually in the form of a program formed from a code. To facilitate version control and contributions between developers, git tools are usually used which are placed on an online platform such as Github. If a repository on an online platform contains sensitive information such as secret key variables and the repository can be accessed by the public, then this can be used to harm the company or organization that owns the repository. Several ways have been done before to detect this secret key variable by using the regex method and using machine learning to minimize false positive results. In this research, we will try to build a text or file classification model that contains secret key variables using the character level CNN and Shannon entropy. The Shannon entropy algorithm is used during pre-processing to retrieve the group of words with the largest entropy value and the CNN character level will be used because it is considered suitable for this case since the value of the secret key is abstract and does not require a sequence context for the said secret key in a file. The results of this study were able to obtain an accuracy value of 96.1% for the accuracy value for the classification of files containing secret key