The development of malware that continues to increase is one of the severe challenges experienced by various sectors, especially in the education sector. Educational institutions that provide open networks and connected to many devices simultaneously become an exciting testing ground for cybercriminals. Anti-virus is often used as prevention by applying signature-based detection of specific tiles contained in the database. Unfortunately, because much new malware continues to grow, the detection of the database becomes inaccurate. Therefore, automatic detection has been carried out using Autoencoder hybridized with ANN and CNN approaches. It will be developed using Python programming language with computer specification of 16GB RAM and Processor i7-7500U. Samples used are malware dataset obtained from various types of repositories such as virussign.com, the Zoo, and certain libraries from researchers. The dimensions of samples will be reconstructed using Autoencoder to generate new values that will be used as input on ANN and CNN. This research also has been developed Principal Component Analysis as a comparison architecture. Those approaches provide outstanding results in detecting malware, scilicet PCA-ANN with an accuracy of 0.985, PCACNN with an accuracy of 0.963, Autoencoder -ANN with an accuracy of 0.994, and Autoencoder -CNN with an accuracy of 0.970. Therefore, those approaches can be used as reference in increasing the need for educational institutions to improve their security from malware by detecting tiles that will be sent or read by each user.
DOI: https://doi.org/10.1109/tale48000.2019.9225899
Publish Year: 2019
